Вы находитесь на странице: 1из 112

Engineering Analysis 2016

Mokin Lee
School of Mechanical Engineering
University of Ulsan

May 11, 2016

Contents

0 Preliminaries

0.1

The Number System . . . . . . . . . . . . . . . . . . . . . . . . . . .

0.2

Equation vs. Identity . . . . . . . . . . . . . . . . . . . . . . . . . . .

0.2.1

Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

0.2.2

Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

0.3.1

Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

0.3.2

Explicit and Implicit Functions . . . . . . . . . . . . . . . . .

0.3.3

Real-valued Functions . . . . . . . . . . . . . . . . . . . . . .

0.3

1 Linear Ordinary Differential Equations

1.1

Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

The Procedures of Solving Linear Ordinary Differential Equations . .

10

1.3

Linear Ordinary Differential Equations of First Order . . . . . . . . .

11

1.4

Linear Independence and the Wronskian . . . . . . . . . . . . . . . .

13

1.5

Linear Homogeneous Differential Equation with Constant Coefficients

15

1.6

Particular Solutions by Method of Undetermined Coefficients . . . . .

17

1.7

Particular Solution by the Method of Variations of Parameters . . . .

19

1.8

Abels Formula for the Wronskian . . . . . . . . . . . . . . . . . . . .

25

1.9

Initial Value Problems and Boundary Value Problems . . . . . . . . .

25

2 Series Solutions of Ordinary Differential Equations

27

2.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

2.2

Power Series Solutions . . . . . . . . . . . . . . . . . . . . . . . . . .

28

2.3

Classification of Singularities . . . . . . . . . . . . . . . . . . . . . . .

33

2.4

Frobenius Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35

3 Special Functions

51

3.1

Bessel Functions

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

3.2

Bessel Function of Order Zero . . . . . . . . . . . . . . . . . . . . . .

54

3.3

Bessel Function of an Integer Order n . . . . . . . . . . . . . . . . . .

55

3.4

Recurrence Relations for Bessel Functions . . . . . . . . . . . . . . .

57

3.5

Bessel Functions of Half Orders . . . . . . . . . . . . . . . . . . . . .

60

3.6

Hankel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

3.7

Modified Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . .

62

3.8

Generalized Equations Leading to Solutions in Terms of Bessel Functions 64

3.9

Zeroes(Roots) of Bessel Functions . . . . . . . . . . . . . . . . . . . .

67

3.10 Graphs of Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . .

68

4 Boundary Value Problems and Eigenvalue Problems


4.1

4.2

74

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

74

4.1.1

Initial Value Problems . . . . . . . . . . . . . . . . . . . . . .

74

4.1.2

Boundary Value Problems . . . . . . . . . . . . . . . . . . . .

76

4.1.3

Eigenvalue Problems . . . . . . . . . . . . . . . . . . . . . . .

77

Examples of Eigenvalue Problems . . . . . . . . . . . . . . . . . . . .

78

4.2.1

Vibration, Wave Propagation and Whirling of Stretched String

78

4.2.2

Longitudinal Vibration and Wave Propagation in Elastic Bars

79

4.2.3

Vibration and Wave Propagation of Beams . . . . . . . . . . .

80

ii

4.3

Adjoint Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81

4.3.1

Self-Adjoint Systems . . . . . . . . . . . . . . . . . . . . . . .

82

4.4

Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . .

84

4.5

Eigenvalue Problems . . . . . . . . . . . . . . . . . . . . . . . . . . .

86

4.6

Properties of Eigenfunctions of Self-Adjoint Systems . . . . . . . . . .

89

4.7

Sturm-Liouville System . . . . . . . . . . . . . . . . . . . . . . . . . .

94

4.8

Sturm-Liouville System for Fourth Order Equations . . . . . . . . . . 101

4.9

Solution of Non-Homogeneous Eigenvalue Problems . . . . . . . . . . 104

iii

Chapter 0
Preliminaries
0.1

The Number System

Natural numbers

Integers 0

Rational numbers

Real numbers

-Negative integers

Complex numbers

Fractional number : 1 , 2 ,

2
7

Irrational numbers : , e, 2,

Imaginary numbers : i = 1

The Natural Numbers


In mathematics, the natural numbers are those used for counting and ordering. Cardinal numbers are used for counting and ordinal numbers are used for ordering.
The natural numbers are closed under addition and multiplication operations.

The Integers
The integer numbers are the combination of the natural numbers, 0, and the negative
integers. Sometimes, the integers are called the whole numbers.
The integer numbers are closed under addition, subtraction, and multiplication operations.
1

The Fractional Numbers


The fractional number numbers can be expressed as the quotient of two integers or
in a fractional form pq of two integers, p and q, with denominator q 6= 0.
The fractional numbers are closed under addition, subtraction, multiplication, and
division operations with a nonzero denominator.

The Rational Numbers


A rational number is any number that can be expressed as the quotient of two integers
or in a fractional form pq of two integers, p and q, with denominator q 6= 0.
The rational numbers x Q are not complete. The rational numbers has gaps on
the corresponding number line.

The Irrational Numbers


Any number that is not rational is called irrational. The irrational numbers include

2, , e, and so on. The irrational numbers x J are not complete. The irrational
numbers has gaps on the corresponding number line.

The Real Numbers


The set of real numbers is complete. The completeness implies that there are not any
gaps or missing points in the real number line. In contrasts, the rational numbers has
gaps on the corresponding number line.
The four basic operations, (+, , , ) are possible in the set of the real numbers.

The Imaginary Numbers

Imaginary number i is defined as i = 1 where i2 = 1, i3 = i, i4 = 1, and so


on. Any imaginary number is defined

x =

xi where x R.

The Complex Numbers


A complex number z is defined
z = a + b i where a, b R.

Notations
1. Natural Numbers: x N
2. Integers: x Z
3. Rational Numbers: x Q
4. Irrational Numbers: x J
5. Real Numbers: x R
6. Imaginary Numbers: x I
7. Complex Numbers: x C

0.2

Equation vs. Identity

Among numerous mathematical expressions, the mathematical expressions containing


the equality sign are the most common expression of the form;
L.H.S = R.H.S
where L.H.S is short for left-handed side and R.H.S is short for right-handed side.
The mathematical equality expression states that the quantity in the left-handed side
is equivalent to the quantity in the right-handed side or vice versa. For example,
2+3=5
An equality expression can contain one or more unknowns.
2+3 = x+1
(x + 1)2 = x2 + 2x + 1
(x + 1)2 = x2 + x + 1
3x2 + y 3 = 1
Depending on how special the unknowns values are, an equality expression can be
classified into identities and equations

0.2.1

Identity

An identity is a mathematical equality statement that always holds true. Examples


are
2+3=5
(x + 1)2 = x2 + 2x + 1

: always true for all x

cos2 + sin2 = 1

: always true for all

e = cos + i sin

: Eulers identity

0.2.2

Equations

In mathematics, an equation is an equality containing one or more variables. In an


equation, the equality is true for some special values of the variable(s).
L.H.S = R.H.S
For example,
2+x=5
2

x + 3x + 2 = 0

true when the unknown variable is x = 3.


true when x = 2 and x = 1.

When an equation is given, we want to know when the equation is true, that is, which
special value(s) of the unknown variable make the equation true?
Hence, our goal is to find the value of an unknown variable that satisfies the given
equation.
There are many types of equations, and they are found in many areas of mathematics.
The techniques used to examine them differ according to their type.

Algebraic equations

Differential equations
Equation

Integral equations

Differo-Integral equations
An equation differs from an identity in that an equation is not necessarily true for all
possible values of the variable.[

Examples
ax + b = 0 (algebraic equation)
ax2 + bx + c = 0 (algebraic equation in explicit form)
x2 + y 2 = a2 (algebraic equation in implicit form)
2

m ddt2x + c dx
+ kx(t) = F (t) (differential equation in explicit form)
dt
y 000 + xy 00 + yy 0 = 0 (differential equation in implicit form)

0.3

Functions

In mathematics, a function is a relation between a set X and a set Y with the property
that each element of the domain is related to exactly one element of the range set.

Set X is called the domain and set Y is called the range or codomain.

0.3.1

Notation

A function f with domain X and range Y is commonly denoted by


f : X Y
For each element x in the domain, the corresponding unique y in the range is called
the function value at x or the image of x under f . It is written as f (x). It can be
said that f associates y with x or maps x to y. This is abbreviated by
y = f (x)
where x is called independent variable and y is called dependent variable.

0.3.2

Explicit and Implicit Functions

A function can be defined by any mathematical condition relating each element in


the domain to the corresponding value in the range. There are many other function
notations.

Explicit Functions
Assuming x is the independent variable and y is the dependent variable, an explicit
function has the form:
y = f (x)
For example,
y = x3 ,

y = sin x,

y = e x + x2

Implicit Functions
Assuming x is the independent variable and y is the dependent variable, an implicit
function has the form:
f (x, y) = 0

or f (x, y) = constant

An implicit form shows a relation between the domain and the range.
For example,
x2 + y 2 = 4,

0.3.3

sin xy = C,

log(x + y) = C

Real-valued Functions

A real-valued function f is one whose range is the set of real numbers or a subset
thereof. If, in addition, the domain is also a subset of the reals, f is a real valued
function of a real variable.
f : X R Y R
or
y = f (x)
where x, y R.

Chapter 1
Linear Ordinary Differential
Equations
1.1

Definitions

A linear ordinary differential equation(LODE) is defined as one that has the following
form:
Ly = a0 (x)

dn y
dn1 y
d2 y
dy
+
a
(x)
+

+
a
(x)
+ an1 (x) + an (x)y = f (x) (1.1)
1
n2
n
n1
2
dx
dx
dx
dx

where
x = the independent variable
y = the dependent variable
f (x) = a known forcing(input) function
a0 (x), a1 (x), , an (x) = given(known) coefficients
Note that the coefficient a0 (x) does not vanish in a x b and a0 (x), a1 (x), , an (x)
are continuous and bounded in a x b.
Define L to be a linear differential operator such as
L = a0 (x)

dn
dn1
d2
d
+
a
(x)
+

+
a
(x)
+ an1 (x) + an (x)
1
n2
n
n1
2
dx
dx
dx
dx

(1.2)

Then, Equation (1.1) can be simply written as


Ly = f (x)

axb

(1.3)

The Order of a Differential Equation


The order of a differential equation is defined as the order of the highest derivative
in the differential equation. Equation (1.1) is an nth order differential equation.

Homogeneous Equation and Homogenous Solution


Equation (1.1) is a homogeneous equation when f (x) = 0. A solution of a homogenous
equation is called the homogeneous solution and is denoted by yh . Then,
Lyh = 0

(1.4)

If a set of n basis functions y1 (x), y2 (x), , yn (x) which are continuous and differential n times, satisfies (), the homogeneous solution to () is obtained by superposition
as
yh (x) = C1 y1 (x) + C2 y2 (x) + Cn yn (x)
where C1 , C2 , , Cn are arbitrary constants.

Non-Homogeneous Equation and Particular Solution


Equation (1.1) is a non-homogeneous equation when f (x) 6= 0. A solution of a nonhomogenous equation is called the particular solution and is denoted by yp . Then,
Lyp = f (x)

(1.5)

A particular solution yp (x) is any solution that satisfies (1.5) and does not contain
any arbitrary constants.

Complete Solution
The complete solution to a differential equation, Ly = f (x) is the sum of the homogeneous solution yh (x) and the particular solution yp (x), that is,
y(x) = yh (x) + yp (x)
= C1 y1 (x) + C2 y2 (x) + Cn yn (x) + yp (x)
where C1 , C2 , , Cn are arbitrary constants.

1.2

The Procedures of Solving Linear Ordinary


Differential Equations

The procedures of solving linear differential equations are summarized as follows:


1. Find the homogeneous solution, by solving the homogeneous differential equation, Lyh = 0.
2. Find the particular solution, by solving the non-homogeneous differential equation, Lyp = f (x).
3. Obtain the complete solution by summing the homogeneous solution and particular solution.

Example:
Solve the following linear ordinary differential equation:
d2 y
+ 4y = 2x2 + 1
2
dx
Solution:
1. The homogeneous solution - yh = C1 sin 2x + C2 cos 2x.
2. The particular solution - yp = x2 /2.
3. The complete solution - y(x) = yh + yp = C1 sin 2x + C2 cos 2x + x2 /2.

We will study how to find a homogeneous solution, a particular


solution and finally the complete solution.

10

1.3

Linear Ordinary Differential Equations of First


Order

Referring to (1.1), a linear ordinary differential equation has the following form:
a0 (x)

dy
+ a1 (x)y = f (x)
dx

(1.6)

where a0 (x) 6= 0 and a1 (x) are continuous and bounded.


Since a0 (x) 6= 0, the following form is obtained by dividing the both sides of
(1.6) by a0 (x) as
dy
+ p(x)y = q(x)
(1.7)
dx
where
a1 (x)
f (x)
p(x) =
and q(x) =
a0 (x)
a0 (x)
We will try to solve (1.7).

1. Homogeneous Solution
The homogenous equation is obtained by assuming q(x) = 0 as
dy
+ p(x)y = 0
dx
The homogenous solution can be obtained by direct integration as
dy
= p(x)y
dx
or

dy
= p(x)dx
y
Integrating the resulting equation gives the homogenous solution:
Z
Z
dy
= p(x)dx + C
y
or

Z
ln y =

p(x)dx + C

Then, the homogenous solution yh (x) is obtained as


yh (x) = e

p(x)dx+C

where C1 is an arbitrary constant.


11

= C1 e

p(x)dx

(1.8)

2. Particular Solution
To obtain the particular solution, one uses an integrating factor (x), such that


dy
d
d
dy
(x)
+ p(x)y =
[(x)y] =
y+
(1.9)
dx
dx
dx
dx
Thus, (x) can be obtained by equating the two sides of (1.9) as follows:
d
= p(x)dx

resulting in a closed form for the integrating factor:


(x) = e

p(x)dx

Using the integrating factor, (1.7) can be rewritten in the form:


d
[(x)yp (x)] = q(x)(x)
dx
or

1
yp (x) =
(x)

q(x)(x)dx = e

p(x)dx

Z
q(x)(x)dx

(1.10)

Complete Solution
The complete solution to (1.1) is obtained by summing (1.8) and (1.10) as
y = yh (x) + yp (x)

= C1 e

p(x)dx

p(x)dx

+e
q(x)(x)dx


Z
R
p(x)dx
= e
C1 + q(x)(x)dx

Any linear ordinary differential equation of first-order has a closed-form


solution!!!.

12

1.4

Linear Independence and the Wronskian

Linear Independence and Linear Dependence


Consider a set of functions [yi (x)], i = 1, 2, . . . , n. A set of of functions are said to be
linear independent on [a, b] if there is no nonvanishing set of constants C1 , C2 , . . . , Cn
which satisfies the following equation identically:
C1 y1 (x) + C2 y2 (x) + + Cn yn (x) = 0

(1.11)

For example, let y1 (x) = x and y2 (x) = 2x and construct the linear combination as

C1 x + C2 x = 0
Then, if C1 = 2 6= 0 and C2 = 1 6= 0 then the linear combination becomes zero.
Hence, the two functions y1 (x) = x and y2 (x) = 2x are linearly dependent.
Another example: let y1 (x) = x and y2 (x) = x2 and construct the linear combination
as
C 1 x + C 2 x2 = 0
Then, then the linear combination becomes zero only if if C1 = 0 and C2 = 0 for all
x. Hence, the two functions y1 (x) = x and y2 (x) = 2x2 are linearly independent.

Wronskian
If y1 (x), y2 (x), , yn (x) satisfy (1.11) and if there exists a set of constants, then
derivatives of (1.11) are also satisfied
C1 y10 (x) + C2 y20 (x) +

+ Cn yn0 (x) = 0

C1 y100 (x) + C2 y200 (x) +

+ Cn yn00 (x) = 0

(n1)

C1 y 1

= 0

= 0

(n1)

(x) + C2 y2

(x) +

13

+ Cn yn(n1) (x) = 0

In matrix form,

y1
y10
y100

(n1)

y1

y2
y20
y200

(n1)

y2

yn

yn0

yn00


(n1)
yn

C1
C2
C3
..
.
..
.
Cn

0
0
0
..
.
..
.
0

(1.12)

For a non-zero set of constants of the homogeneous algebraic equation (1.12), the
determinant of the coefficient of C1 , C2 , . . . , Cn must vanish. The determinant
W (y1 , y2 , . . . , yn )



y1
y

y
2
n




0
0
0
y
y

y


1
2
n


00
00

y100

y
y
n
2

W (y1 , y2 , . . . , yn ) =







(n1) (n1)
(n1)
y
y
yn
1

is referred to as the Wronskian1 of y1 , y2 , . . . , yn .


If the Wronskian of a set of unction is not identically zero, the set of function
[yi (x)] is linearly independent set. The nonvanishing of the Wronskian is a necessary
and sufficient condition for linear independence of [yi (x)] for all x.
Example: Let y1 (x) = x and y2 (x) = 2x Then, the Wronskian is evaluated as


x 2x



= 2x 2x = 0
1 2
Hence, the two functions y1 (x) = x and y2 (x) = 2x are linearly dependent.
Example: Let y1 (x) = x and y2 (x) = x2 Then, construct the Wronskian as


x x2


x.

= 2x2 x2 = x2 6= 0
1 2x
Hence, the two functions y1 (x) = x and y2 (x) = x2 are linearly independent.
1

Jozef Maria Hoene-Wronski, Polish mathematician (17761853)

14

1.5

Linear Homogeneous Differential Equation


with Constant Coefficients

Consider the linear homogeneous differential equation of order n with constant coefficients:
dn y
dn1 y dn2 y
d2 y
dy
+ an y = 0
+
a
a
+

+
a
+ an1
1
2
n2
n
n1
n2
2
dx
dx
dx
dx
dx
where a0 , a1 , , an1 , an are constants, with a0 6= 0 can be readily solved.
a0

(1.13)

Since function ex can be differentiate many times, the we may try:


y = ex
where is a constant.
Substituting (1.5) into (1.13) results in

Ly = a0 n + a1 n1 + + an2 2 + an1 + an ) y = 0
The resulting polynomial equation of degree n:
a0 n + a1 n1 + + an2 2 + an1 + an ) = 0

(1.14)

is called the characteristic equation.


(i) n Distinct Roots of the Characteristic Equation
If the polynomial in (1.14) has n distinct roots, 1 , 2 , , n , then there are distinct
n solutions of the form. Then, the characteristic equation can be factored out as
a0 ( 1 ) ( 2 ) ( n ) = 0

(1.15)

where 1 6= 2 6= 6= n are constants.


Then, there are n linearyly independent solutions to (1.13):
yi (x) = ei x ,

, i = 1, 2, . . . , n

(1.16)

each of which satisfies (1.13).


The general form of the homogeneous equation (1.13) can be written in terms
of the n independent solutions of (1.16):
yh = C1 e1 x + C2 e2 x + + Cn en x
where Ci are arbitrary constants.
15

(ii) Repeated Roots of the Characteristic Equation


If there are k repeated roots, for example the j th root is repeated k times, then there
are nk +1 independent solution and a method must devised to obtain the remaining
k 1 solutions.
The characteristic equation in (1.14) can be rewritten as follows:
a0 n + a1 n1 + + an2 2 + an1 + an
k times

}|
{
z
= a0 ( 1 ) ( 2 ) ( j1 ) ( j ) ( j ) ( j ) ( j+k ) ( n )
= a0 ( 1 ) ( 2 ) ( j1 ) ( j+k ) ( n ) ( j )k

(1.17)

To obtain the missing solution, a solution of the form xm ej x is substituted into (1.13)
and the total homegeneous solution is obtained as

yh (x) =C1 e1 x + C2 e2 x + + Cj + Cj+1 x + Cj+2 x2 + + Cj+k1 xk1 ej x
+ Cj+k ej+l x + + Cn en x

16

1.6

Particular Solutions by Method of Undetermined Coefficients

Particular solutions to general nth order linear differential equation can be obtained
by the method of variation of parameters. However, there are simple means for
obtaining particular solutions to non-homogenous differential equations with constant
coefficients:
Table 1.1: Trial Functions

(a)
(b)
(c)
(d)

f (x)
sin ax or cos ax
ex
sinh ax or cosh ax
xm

try yp
yp = A sin ax + B cos ax
yp = Cex
yp = A sinh ax + B cosh ax
yp = F0 xm + F1 xm1 + + Fm1 x + Fm

If f (x) is a product of the function given in (a)(d), then a trial solution can be
written in the form of the product of the corresponding trial solutions. For example,
if
f (x) = x2 e2x sin 3x
then one uses a trial particular function as
yp = F 0 x 2 + F 1 x + F 2


e2x (A sin 3x + B cos 3x)

Note that if a factor or term of f (x) happens to be one of the solutions of the
homogeneous solutions, then the portion of the trial solution yp corresponding to that
term or factor of f (x) must be multiplied by xk , where an integer k is chose such that
the portion of trial solution is one power of x higher than any of the homogeneous
solution.
Example:
d2
d3

3
+ 4y = 40 sin 2x + 27x2 ex + 18xe2x
3
2
dx
dx
Solution: the homogeneous solution is obtained by
yh = C1 ex + (C2 + C3 x)e2x
17

(a)
(b)
(c)

f (x)
sin 2x
x2 ex
xe2x

try yp
yp = A sin 2x + B cos 2x
yp = (Cx2 + Dx + E)xex
yp = (F x + G)x2 e2x

For the particular solution, try Thus, the trial particular solution becomes
yp = A sin 2x + B cos 2x + Cx2 + Dx + E)xex + (F x + G)x2 e2x
Substitution of yp into the differential equation equating the coefficients, one obtains:
A = 2 B = 1 C = 1 D = 2 E = 2 F = 1 G = 1
Then, the particular solution is obtained as
yp = 2 sin 2x + cos 2x + (x2 + 2x + 2)xex + (x 1)x2 e2x
Thus, the complete solution is obtained by
y = C1 ex + (C2 + C3 x)e2x + 2 sin 2x + cos 2x + (x3 + 2x2 + 2x)ex + (x3 x2 )e2x

18

1.7

Particular Solution by the Method of Variations of Parameters

Except for differential equations with constant coefficients, it is very difficult to find
the particular solution. The method of undetermined coefficient is restricted to linear
ordinary differential equations with constant coefficients, where input function f (x)
is an elementary function listed in Table (1.1) As a general method, the method of
variations of parameters is used. This method is credited to Joseph Lagrange 2 .
The homogeneous differential equation (1.1) has n independent solutions, i.e.,
yh = C1 y1 + C2 y2 + + Cn yn
Assume that particular solution yp of (1.5) can be obtained from n products of these
solutions with n unknown functions v1 (x), v2 (x), , vn (x), i.e.,
yp = v1 y1 + v2 y2 + + vn yn

(1.18)

Differentiating (1.18) results in:


yp0 = (v10 y1 + v20 y2 + + vn0 yn ) + (v1 y10 + v2 y20 + + vn yn0 )

(1.19)

Since yp in (1.18) must satisfy one equation (1.1), one can arbitrarily specify (n 1)
more relationships. Thus, let
v10 y1 + v20 y2 + + vn0 yn = 0

(1.20)

yp0 = v1 y10 + v2 y20 + + vn yn0

(1.21)

so that
Differentiating yp0 once again gives:
yp00 = (v10 y10 + v20 y20 + + vn0 yn0 ) + (v1 y100 + v2 y200 + + vn yn00 )

(1.22)

v10 y10 + v20 y20 + + vn0 yn0 = 0

(1.23)

yp00 = v1 y100 + v2 y200 + + vn yn00

(1.24)

Again let:
resulting in
Carrying this procedure to the (n 1)st derivative, one obtains


 
(n2)
(n2)
(n1)
(n1)
yp(n1) = v10 y1
+ v20 y2
+ + vn0 yn(n2) + v1 y1
+ v2 y2
+ + vn yn(n1)
(1.25)
2

French mathematician, (17361813)

19

and letting
(n2)

v10 y1

(n2)

+ v20 y2

+ + vn0 yn(n2) = 0

(1.26)

then
(n1)

yp(n1) = v1 y1

(n1)

+ v2 y 2

+ + vn yn(n1)

(1.27)

Thus far (n 1) conditions have been specified on the functions v1 , v2 , , vn . The


nth derivative is obtained in the form


 
(n)
(n)
(n)
0 (n1)
0 (n1)
0 (n1)
(n)
+ v1 y1 + v2 y2 + + vn yn
+ + vn yn
+ v2 y2
yp = v1 y1
(1.28)
Substituting of the solution yp in (1.18) and its derivatives in (1.21), (1.24), (1.27),
(1.28) into (1.1),
a0 (x)yp(n) + a1 (x)yp(n1) + + an1 (x)yp0 + an (x)yp
h
 
i
(n)
(n)
(n)
0 (n1)
0 (n1)
0 (n1)
= a0 (x) v1 y1 + v2 y2 + + vn yn + v1 y1
+ v2 y2
+ + vn yn


(n1)
(n1)
+ a1 (x) v1 y1
+ v2 y2
+ + vn yn(n1)


(n2)
(n2)
(n2)
+ a2 (x) v1 y1
+ v2 y2
+ + vn yn
+
+ an1 (x) (v1 y10 + v2 y20 + + vn yn0 )
+ an (x) (v1 y1 + v2 y2 + + vn yn ) = f (x)
and grouping together derivatives of each solution, one obtains


(n1)
(n2)
v1 a0 y1n + a1 y1
+ a2 y1
+ + an1 y10 + an y1


(n1)
(n2)
n
0
+ v2 a0 y2 + a1 y2
+ a2 y2
+ + an1 y2 + an y2


(n1)
(n2)
+ v3 a0 y3n + a1 y3
+ a2 y3
+ + an1 y30 + an y3
+
+ vn a0 ynn + a1 yn(n1) + a2 yn(n2) + + an1 yn0 + an yn


(n1)
(n1)
+ a0 (x) v10 y1
+ v20 y2
+ + vn0 yn(n1) = f (x)


(1.29)

Since Ly1 = 0, Ly2 = 0, ,Lyn = 0, (1.29) becomes




(n1)
(n1)
+ + vn0 yn(n1) = f (x)
a0 (x) v10 y1
+ v20 y2
or
(n1)

v10 y1

(n1)

+ v20 y2

+ + vn0 yn(n1) =
20

f (x)
a0 (x)

(1.30)

Collecting (1.20), (1.23), (1.26), (1.27),and (1.30), the system algebraic equations on
the unknown functions v10 , v20 , , vn0 can now be written as follows:
v10 y1 + v20 y2 + + vn0 yn = 0
v10 y10 + v20 y20 + + vn0 yn0 = 0

(n2)

+ + vn0 yn(n2) = 0

(n1)

+ + vn0 yn(n1) =

(n2)
v10 y1

+ v20 y2

(n1)

+ v20 y2

v10 y1

f (x)
a0 (x)

In matrix form,

y1
y2

0
y20
y1

y100
y200

(n2) (n2)
y1
y2
(n1)
(n1)
y1
y2

yn

yn0

yn00

(n2)
yn
(n1)
yn

0
v1

v0
2
..
.

..
.

v0
n1
vn0

0
0
..
.
..
.
0
f (x)
a0 (x)

Since the homogeneous solutions, y1 (x), y2 (x), . . . , yn (x) are linearly independent, the
Wronskian of the solution is not zero, that is,
W (y1 (x), y2 (x), . . . , yn (x)) 6= 0
Then, the unknown functions v10 , v20 ,
rule as

0
y2

0
y20


0
y200



(n2)
0
y2
f (x)
(n1)

y2
a0 (x)
0
v1 =
y1
y2


0
y20
y1

y100
y200


(n2) (n2)
y1
y
(n1) 2(n1)
y
y2
1
21

, vn0 can now be written by using Cramers

















yn

yn0

yn00

(n2)

yn

(n1)
yn

yn

yn0

yn00

(n2)
yn
(n1)
yn


y1
0

y0
0

1

00
y1
0


(n2)
y1
0
(n1) f (x)
y
1
a0 (x)
v20 =
y1
y2


0
y20
y1

y100
y200


(n2) (n2)
y1
y
(n1) 2(n1)
y
y2
1
















yn

yn0

yn00

(n2)

yn

(n1)
yn

yn

yn0

yn00

(n2)
yn
(n1)
yn

=
=
=
=
=
=


y1
y2

0
yn

y0

0
yn0
y20

1


y100

0
yn00
y200





(n2) (n2)
(n2)
y1

y

0
yn
(n1) 2(n1)

(n1)
f (x)
y
y

y
n
1
2
a0 (x)
0
=
vn1

y1
y2

yn



y20

yn0
y10


y100
y200

yn00




(n2) (n2)
(n2)
y1

y
yn
(n1) 2(n1)

(n1)
y
y

y
n
1
2

22


y1

y0

1

y100



(n2)
y1
(n1)
y
1
0
vn =
y1


y10

y100



(n2)
y1
(n1)
y
1

y2
y20
y200

(n2)

y2
(n1)
y2
y2
y20
y200

(n2)

y2
(n1)
y2

yn1
0
yn1
00
yn1

(n2)
yn1
(n1)
yn1

0
0
0

0
f (x)
a0 (x)

yn

yn0

yn00

(n2)
yn
(n1)
yn

Equations give a unique closed-form solution for v10 , v20 , , vn0 , which can be integrated to give v1 , v2 , , vn as
Z x
Z x
Z x
0
0
v1 (x) =
v1 ()d, v2 (x) =
v2 ()d, , vn (x) =
vn0 ()d
Therefore, a particular solution yp (x) can be obtained, i.e.,
yp = v1 y1 + v2 y2 + + vn yn
where v1 , v2 , , vn are a set of homogeneous solutions.

Once a set of homogenous basis functions is obtained, the particular solution can be found by using Method of Variations of
Parameters.
Hence, a key to solving linear ordinary differential equations is to
find a set of homogenous solutions.

23

The method of variation of the parameters is applied to a general 2nd order


differential equation. Let:
a0 (x)y 00 + a1 (x)y 0 + + a2 (x)y = f (x)
such that the homogeneous solution is given by
yh = C1 y1 (x) + C2 y2 (x)
and a particular solution can be found in the form:
yp = v1 (x)y1 (x) + v2 (x)y2 (x)
The functions v10 and v20 can be found by solving two algebraic equations:
v10 y1 + v20 y2 = 0
f (x)
v10 y10 + v20 y20 =
a0 (x)
Solving for v10 and v20 , one obtains
y2 f (x)/a0 (x)
y2 f (x)
=
0
0
y1 y2 y1 y2
a0 (x)W (x)
y1 f (x)
y1 f (x)/a0 (x)
=+
=
0
0
y1 y2 y1 y2
a0 (x)W (x)

v10 =
v20

Direct integration of these two expressions gives:


Z x
y2 f ()
v1 (x) =
d
a0 ()W ()
Z x
y1 f ()
v2 (x) = +
d
a0 ()W ()
The unknown functions v1 (x) and v2 (x) are then substituted into yp to give:
Z x
Z x
y2 f ()
y1 f ()
yp = y1 (x)
d + y2 (x)
d
a0 ()W ()
a0 ()W ()

24

1.8

Abels Formula for the Wronskian

The Wronskian can be obtained in a closed form when the set of functions
y1 , y2 , , yn are solutions of an ordinary differential equation:
dW
a1 (x)
=
W
dx
a0 (x)
which can be integrated to give a closed form formula for the Wronskian:
W (x) = W0 e

a1 (x)
dx
a0 (x)

with W0 = constant. This is known as Abels Formula3 .

1.9

Initial Value Problems and Boundary Value


Problems

For a unique solution of an ordinary differential equation of order n, whose complete


solution contains n arbitrary constants, a set of of n conditions on the dependent
variable is required.
y = yh (x) + yp (x)
= C1 y1 (x) + C2 y2 (x) + + Cn yn (x) + yp (x)
where [Ci ] are arbitrary constants and [yi (x)] are a set of basis functions of the
homogegenous equation.

Initial Value Problems


The set of n-conditions on the dependent variable is a set of the values that the
dependent variable and its first (n 1) derivatives take at a point x = x0 , which can
be given as
y(x0 ) = 0
y 0 (x0 ) = 1
..
.
. = ..

a x, x0 b

y (n2) (x0 ) = (n2)


y (n1) (x0 ) = (n1)
3

Niels Henrik Abel- Norwegian mathematician, 18021829

25

A unique solution for the set of constants [Ci ] in the homogeneous solution yh (x) can
be determined by using the conditions specified at a point x = x0 . Such problems are
known are Initial Value Problems.

Boundary Value Problems


If the set of n-conditions on the dependent variable is specified on two end points of
a bounded region valid in the closed region between the two end point, these points
are called Boundary Points and the conditions specified on the dependent variable
and its derivatives up to the (n 1)st are called Boundary Conditions(BC). For
example, the boundary conditions are given by
y(a) = 0
y 0 (b) = 1
..
.
. = ..

axb

y (n2) (a) = (n2)


y (n1) (b) = (n1)
Such problems are referred to as Boundary Value Problems.

26

Chapter 2
Series Solutions of Ordinary
Differential Equations
2.1

Introduction

In many cases, it is not possible to obtain the solution of an ordinary differential


equation of the type of (2.1) in a closed form:
Ly = a0 (x)

dn y
dn1 y
d2 y
dy
+
a
(x)
+

+
a
(x)
+ an1 (x) + an (x)y = 0 (2.1)
1
n2
n
n1
2
dx
dx
dx
dx

If a0 (x) 6= 0, a1 (x), a2 (x), , an (x) are continuous and bounded in the interval
a x b, then there exists a set of n solutions yi (x), i = 1, 2, . . . , n. Such a solution
can be expanded into a Taylor series about a point x0 , a < x0 < b such that
y(x) =

cn (x x0 )n

(2.2)

n=0

where

y (n) (x0 )
cn =
n!

(2.3)

The Taylor series expression in (2.2) is a power series about the point x = x0 .
In general, one does not know y(x) beforehand, so that the coefficients of the series
cn are not determinable from (2.3). However, one can assume that the solution to
(2.1) has a power series and then the unknown coefficients cn can be determined by
substituting (2.2) into (2.1).

27

2.2

Power Series Solutions

Power series solutions about x = x0 of the form in (2.2) can be transformed to a


power series solution about z = 0.
Letting z = x x0 , (2.2) transforms to
a0 (z+x0 )

dn y
dn1 y
d2 y
dy
+a
(z+x
)
+

+a
(z+x
)
+an1 (z+x0 ) +an (z+x0 )y = 0
1
0
n2
0
n
n1
2
dz
dz
dz
dz

Thus, power series homogeneous solutions about x = x0 become series solutions


about z = 0, i.e.,

X
cm z m
y(z) =
m=0

Hence, one need to discuss power series solutions about the origin, which will be taken
to be x0 = 0 for simplicity, i.e.,
y(x) =

cm x m

(2.4)

m=0

Substitution of the series in (2.4) into (1.1) and equation the coefficients of each power
of x to zero, results in an infinite number of algebraic equations, each one gives the
constant cm in terms of cm1 , cm2 , . . . , c1 , and c0 for m = 1, 2, . . ..
Since the homogeneous differential equation is of order n, then there will be n
arbitrary constants, i.e., the constants c0 , c1 , . . . , cn are arbitrary constants. The constants cn+1 , cn+2 , . . . can be computed in terms of the arbitrary constants c0 , c1 , . . . , cn .

28

Example:
Obtain the power series solution valid in the neighborhood of x0 = 0 of the following
equation:
d2 y
xy = 0
dx2
Since a0 (x) = 1 6= 0, a1 (x) = 0, and a2 (x) = x , all bounded, there exists a solution
to this linear ordinary differential equation.
Let the solution to be in the form of a power series about x0 = 0.
y=

cn x

y =

n=0

ncn x

00

n1

y =

n=0

n(n 1)cn xn2

n=0

Substituting the power series and its derivatives into the differential equation gives:
Ly =

n(n 1)cn x

n=0

n2

cn xn+1

n=0

Writing out the two series in a power series of ascending power of x results in:
0 c0 x2 + 0 c1 x1 + 2c2 + (6c3 c0 )x + (12c4 c1 )x2
+ (20c5 c2 )x3 + (30c6 c3 )x4 + (42c7 c4 )x5 + = 0
Since the power series of all null function has zero coefficients, then equating the
coefficient of each power of x to zero, one obtains:
c0 =
c3 =
c6 =

0
= indeterminate
0
c0
c0
= 23
6
c3
c0
= 2356
65

c1 =
c4 =
c7 =

0
= indeterminate
0
c1
c1
= 34
12
c4
c0
= 3467
67

c2 = 0
c2
c5 = 5 cot
4

Thus, the series solution becomes:


c0 3
c1 4
c0
c1
y = c0 + c1 x +
x +
x +
x6 +
x7 +
2

3
3

4
2

6
3

7




x3
x6
x4
x7
= c0 1 +
+
+ + c1 x +
+
+
6
6 30
12 12 42
Since c0 and c1 are arbitrary constants, then
y1 (x) = 1 +

x6
x3
+
+
6
6 30

and

x4
x7
+
+
12 12 42
are the two independent solutions of the homogeneous differential equation.
y2 (x) = x +

29

Recurrence Formula
It is more advantageous to work out the relationship between cn and cn1 , cn2 , . . . , c1 , c0
in a formula known as the Recurrence Formula.
Assuming the solution and its derivatives in a power series as
y=

cn x n

n=0

y0 =

ncn xn1

n=0

y 00 =

n(n 1)cn xn2

n=0

Rewriting Ly = 0 in expanded form and separating the first few terms of each series,
such that the remaining terms of each series start at the same power of x1 gives when
substituted into the differential equation:
Ly =

n(n 1)cn x

n2

cn xn+1

n=0

n=0

= 0 c0 x2 + 0 c1 x1 + 2 1 c2 x0 +

n(n 1)cn xn2

cn xn+1 = 0

n=0

n=3

where the first term of each power series starts with x1 .


Letting k = n 3 in the first series and k = m in the second series, so that the
two series start with the same index m = 0 and the power of x is the same for both
series, one obtains:
Ly =

n(n 1)cn x

n2

n=0

cn xn+1

n=0

= 0 c0 x2 + 0 c1 x1 + 2 1 c2 x0 +

n(n 1)cn xn2

n=3

cn xn+1

n=0

X
X
2
1
0
k+1
= 0 c0 x + 0 c1 x + 2 1 c2 x +
(k + 3)(k + 2)ck+3 x

ck xk+1
k=0

= 0 c0 x2 + 0 c1 x1 + 2 1 c2 x0 +

k=0

[(k + 3)(k + 2)ck+3 ck ] xk+1 = 0

k=0

For the above equality to hold, one obtains


c0 = indeterminate

c1 = indeterminate
30

c2 = 0

and

[(k + 3)(k + 2)ck+3 ck ] xk+1 = 0

k=0

Equating the coefficient of xk+1 to zero gives the recurrence formula:


ck+3 =

ck
(k + 2)(k + 3)

k = 1, 2, 3, . . .

This recurrence formula relates ck+3 to ck and results in the same constants evaluated
earlier.
The recurrence formula reduces the amount of algebraic manipulations needed
for evaluating the coefficients cm .
Example:
Solve the following ordinary differential equation about x0 = 0
x

dy
d2 y
+ 3 + xy = 0
2
dx
dx

Since a0 (x) = x 6= 0, a1 (x) = 3, and a2 (x) = x , all bounded, there exists a solution
to this linear ordinary differential equation.
Let the solution to be in the form of a power series about x0 = 0.
y=

cn x

y =

ncn x

00

n1

y =

n(n 1)cn xn2

n=0

n=0

n=0

Substituting into the differential equation gives:


Ly =

n(n 1)cn xn1 + 3

n=0

ncn xn1 +

n=0

n(n + 2)cn xn1 +

n=0

= 0 c0 x1 + 3 c1 +

cn xn+1

n=0

cn xn+1

n=0

n=2

n=0

n(n + 2)cn xn1 +

cn xn+1 = 0

Substituting k = n 2 in the first and k = n in the second series, one obtains


Ly = 0 c0 x

X
X
k+1
+ 3 c1 +
(k + 2)(k + 4)c( k + 2)x
+
ck xk+1

= 0 c0 x1 + 3 c1 +

k=0

k=0



(k + 2)(k + 4)c( k + 2) + ck xk+1 = 0

k=0

31

Thus, equating the coefficient of each power of x to zero gives:


c0 = indeterminate

c1 = 0

as well as the recurrence formula:


ck+2 =

ck
,
(k + 2)(k + 4)

k = 0, 2, 4, 6, . . .

The recurrence formula can be used to evaluate the remaining coefficients:


c1
c2
0
c3 = 15
= 0 c4 = 24
=
c2 = 22c2!1!
c3
0
c5 = 35
= 0 = c7 = c9 = c6 = 26c4!2!

c0
24 3!2!

Thus, the series solution obtainable in the form of a power series is:


x4
x6
x2
+

+
y = c0 1 2
2 2!1! 24 3!2! 26 4!2!
Since c0 is a arbitrary constant, then
y1 = 1

x2
x4
x6
+

+
22 2!1! 24 3!2! 26 4!2!

This solution has only one arbitrary constant, thereby giving one solution. The
missing second solution cannot be obtained in a power series form due to the fact
that a0 (x) = x vanishes at the point about which the series is expanded, i.e., x = 0 is
a singular point of the differential equation. To obtain the full solution, one needs to
deal with differential equations having singular points at the point of expansion x0 .

32

2.3

Classification of Singularities

Dividing the second order differential equation by a0 (x), then it becomes


d2 y
dy
2 (x)y = 0
+a
1 (x) + a
2
dx
dx
where a
1 (x) = a1 (x)/a0 (x) and a
2 (x) = a2 (x)/a0 (x).
Ly =

(2.5)

If either of the two coefficients a


1 (x) or a
2 (x) are unbounded at a point x0 , then
the equation has a singularity at x = x0 .
(i) If a
1 (x) or a
2 (x) are bounded(regular) at x0 , then x0 is called a regular
point(RP).
(ii) If x = x0 is a singular point and if

lim
(x

x
)
a
(x)

finite
xx
0
1
0

x0 is a Regular Singular Point(RSP)


and

2
limxx0 (x x0 ) a
2 (x) finite
(iii) If x = x0 is a singular point and if either

a1 (x) unbounded
limxx0 (x x0 )

x0 is a Irregular Singular Point(ISP)


or

2
limxx0 (x x0 ) a
2 (x) unbounded

Example
Classify the behavior of each of the following differential equations at x = 0 and at
all the singular points of each equation.
(a)
d2 y
dy
+
sin
x
+ x2 y = 0
dx2
dx
where a
1 (x) = sinx x and a
2 (x) = x.
Both coefficients are regular at x = 0, thus x = 0 is a Regular Point.
x

(b)
x

d2 y
dy
+
3
+ xy = 0
dx2
dx

where a
1 (x) = x3 and a
2 (x) = 1.
Point x = 0 is a singular point. Classifying the singularity at x = 0
lim x

xx0

3
=3
x

lim x2 (1) = 0

xx0

33

Thus, x = 0 is a regular singular point(RSP).


(c)
x2 (x2 1)

d2 y
dy
+ x2 y = 0
+ (x 1)2
2
dx
dx

1
where a
1 (x) = x(x1)
2 (x) = (x1)(x+1)
.
2 (x+1) and a
Point x = 0 is a singular point. Classifying the singularity at x = 0

lim x

x0

3
=3
x

lim x2 (1) = 0

x0

There are three singular points: x = 1, 0, 1.


x0 = 1

(x1)

limx1 (x + 1) x2 (x+1) = 2

limx1 (x +

(x1)
1)2 (x1)(x+1)

= 0

x0 = 1 is a RSP.

x0 = 0

(x1)

limx0 x x2 (x+1) =

(x1)
limx0 x2 (x1)(x+1)

= 0

x0 = 0 is an ISP.

x0 = +1

(x1)

limx+1 (x 1) x2 (x+1) = 0

limx+1 (x 1)

1
(x1)(x+1)

= 0

34

x0 = +1 is a RSP.

2.4

Frobenius Solution

If a differential equation has a Regular Singular point at x0 , then one or both solution(s) may not be obtainable by the power series expansion.
If the equation has a singularity at x = x0 , one can perform a linear transformation, and seek a solution about z0 . For simplicity, a solution valid in the neighborhood
of x = 0 is presented.
For equations that have a RSP at x = x0 , a solution of the form:

X
y(x) =
an (x x0 )n+

(2.6)

n=0

can be used, where is an unknown constant to be determined later. This solution


is known as the Frobenius Solution1 .
Since a
1 (x) and a
2 (x) can, at most, be singular to the order of (x x0 )1 and
(x x0 )2 , then
(x x0 )
a1 (x)

(x x0 )2 a
2 (x)

and

are regular functions in the neighborhood of x = x0 . Thus, expanding the above


functions into a power weries about x = x0 results in:

X
2
k (x x0 )k
(2.7)
(x x0 )
a1 (x) = 0 + 1 (x x0 ) + 2 (x x0 ) + =
k=0

and
(x x0 )2 a
2 (x) = 0 + 1 (x x0 ) + 2 (x x0 )2 + =

k (x x0 )k

(2.8)

k=0

Transforming the equation by z = x x0 and replacing z by x, one can discuss


solutions about x0 = 0. The Frobenius solution in (2.6) and series expansions of
a
1 (x) and a
1 (x) about x0 = 0 in (2.7) and (2.8) are substituted into the homogeneous
differential equation (2.5), such that
Ly =

(n + 1)(n + )an xn+2

n=0

k xk1

!"
X

(n + )an xn+1

n=0

k=0

X
k=0

k xk2

!"
X

#
(n + )an xn+ = 0 (2.9)

n=0

Ferdinand Georg Frobenius, German mathematician (1849-1917)

35

The second term in (2.9) can be written in a Taylor series form as follows:
!"
#
"

X
X
k1
n+1
2
k x
(n + )an x
=x
0 a0 + [0 a0 + ( + 1)1 a0 ] x
n=0

k=0

+ [a0 2 + ( + 1)1 a1 + ( + 2)2 a0 ] x2 + +

( k=n
X

)
( + k)ak nk

#
xn +

k=0

cn xn+2

n=0

where
cn =

k=n
X

( + k)ak nk

k=0

The third term in (2.9) can be written in a Taylor series form as follows:
!"
#

X
X
X
k xk2
an xn+ =
dn xn+2
k=0

n=0

where
dn =

n=0

k=n
X

ak nk

k=0

Equation (2.9) becomes then


#
"

k=n
X
X
X
cn xn+2
( + k)ak nk +
Ly = x2
(n + 1)(n + )an xn+2 +
n=0

=x

k=0

n=0

{( 1) + 0 + 0 } a0 + {( + 1) + ( + 1)0 + (1 + 1 )a0 } x +
(2.10)

Defining the quantities:


f () = ( 1) + 0 + 0
fn () = n + n
then, (2.10) can be rewritten in a condensed form:
h
Ly = x2 f ()a0 + {f ( + 1)a1 + f1 ()a0 } x + {f ( + 2)a2 + f1 ()a1 + f2 ()a0 } x2
i
n
+ + {f ( + n)an + f1 ( + n 1)an1 + + fn ()} x +
"
(
) #

X
X
= x2 f ()a0 +
f ( + n)an +
fk ( + n k)ank xn
(2.11)
n=1

k=1

36

Each of the constants a1 , a2 , . . . , an , can be written in terms of a0 by equating


the coefficients of x, x2 , . . . to zero as follows:
f1 ()
a0
f ( + 1)
f1 ()a1 + f2 ()a0
a2 () =
f ( + 2)
g2 ()
f1 ()f ( + 1) + f2 ()f ( + 1)
a0 =
a0
=
f ( + 1)f ( + 2)
f ( + 2)
f1 ( + 2)a2 + f2 ( + 1)a1 + f3 ()a0
g3 ()
a0 =
a0
a3 () =
f ( + 3)
f ( + 3)
a1 () =

and by induction:
an () =

gn ()
a0 ,
f ( + n)

n1

(2.12)

Substitution of an (), n = 1, 2, 3, . . . in terms of the coefficient a0 into (2.11) results


in the following expression for the differential equation:
Ly = x2 f ()a0 = 0

(2.13)

and consequently the series solution can be written in terms of an (), which is a
function of () and a0 :

X
y(x, ) = a0 x +
an xn+
(2.14)
n=1

For a non-trivial solution; a0 6= 0, (2.14) is satisfied if:


f () = ( 1) + 0 + 0 = 0

(2.15)

Equation (2.15) is called the Characteristic Equation, which has two roots 1 and 2 .
Depending on the relationship of the two roots, there are three different cases.

Case (a): Two roots are distinct and do not differ by an integer:
If 1 6= 2 and 1 2 6= integer, then there exists two solutions to (2.9) of the form:
y1 (x) =

an (1 )xn+1

(2.16)

an (2 )xn+2

(2.17)

n=0

and
y2 (x) =

X
n=0

37

Case (b): Two identical roots 1 = 2 = 0 :


If 1 = 2 = 0 , then only one possible solution can be obtained by the method
of Case (a), (2.16), i.e.,

X
y1 (x) =
an (0 )xn+0
(2.18)
n=0

where a0 = 1.
To obtain the second solution, one must use (2.13) and (2.14). If 1 = 2 = 0 ,
then the characteristic equation has the form:
f () = ( 0 )2
and (2.13) becomes:
Ly(x, ) = x2 ( 0 )2 a0

(2.19)

where y(x, ) is given in (2.14).


First differentiate (2.19) partially with :



y(x, )
Ly = L
= a0 2( 0 ) + ( 0 )2 log x x2

where

d
x = x log x
d

If = 0 , then:

y(x, )
L


=0
=0

Thus, the second solution satisfying the homogeneous differential equation is given
by:
y(x, )
.
y2 (x) =

=0
Using the form of the Frobenius solution:

y(x, ) = a0 x +

an ()xn+

n=1

then differentiating the expression for y(x, ) with results in:

X
X
y(x, )

0
n+
= a0 x log x +
an ()x
+
an ()xn+ log x

n=1
n=1

= log x

an ()x

n+

n=0

X
n=1

38

a0n ()xn+

where a0n () =

dan ()
.
d

Thus, the second solution for the case of equal roots takes the form with a0 = 1:
y2 (x) = log x

an ()x

n+

n=0

a0n ()xn+

n=1

= y1 (x) log x +

a0n (0 )xn+0

(2.20)

n=1

Case (c): Distinct roots that differ by an integer:


If 1 2 = k, a positive integer, then characteristic equation becomes
f () = ( 1 )( 2 ) = ( 2 k)( 2 ),

1 > 2

First, one can obtain the solution corresponding to the larger root = 1 in the
form given in (2.16)

X
y1 (x) =
an (1 )xn+1
n=0

where a0 = 1.
The second solution corresponding to the smaller root = 2 may have the
coefficient ak (2 ) unbounded, because from (2.12), the expression for ak (2 ) is:

gk ()
ak (2 ) =

f ( + k)
=2

where the denominator vanishes at = 2 :


f ( + k)|=2 = ( + k 2 k)( + k 2 )|=2
= ( 2 )( + k 2 )|=2 = 0
Thus, unless the numerator gk (2 ) also vanishes, the coefficient ak (2 ) becomes unbounded.
If gk (2 ) vanishes, then ak (2 ) is indeterminate and one may start a new infinite
series with ak , i.e.,


k1or
X
X  an (2 ) 
an (2 ) n+2
n+2
x
+ ak
x
y2 (x) = a0
a0
ak
n=0
n=k

k1or

X  an (2 ) 
X
am+k (2 ) m+1
n+2
= a0
x
+ ak
x
(2.21)
a0
ak
n=0
m=0
39

It can be shown that solution preceded by the constant ak is identical to y1 (x), thus
one can set ak = 0 and a0 = 0. The first part of the solution with a0 may be finite
polynomial or an infinite series. depending on the order of the recurrence formula
and on the integer k.
If gk (2 ) does not vanish, then one must find another method to obtain the
second solution. A new solution similar to Case (b) is developed by removing the
constant 2 from the denominator of an (k). Since the characteristic equation in
(2.13) is given by;
Ly(x, ) = a0 x2 f () = a0 x2 ( 1 )( 2 ) = 0

(2.22)

then, multiplying (2.22) by ( 2 ) and differentiating partially with , one obtains:





[( 2 )Ly] =
[L( 2 )y(x, )] = L
( 2 )y(x, )


 2
= a0
x ( 1 )( 2 )2



= a0 ( 1 )( 2 )2 x2 log x + x2 ( 2 )2 + 2 x2 ( 1 )( 2 )
Thus, the function that satisfies the homogeneous differential equation:



( 2 )y(x, )
=0
L

=2
gives an expression for the second solution, i.e.,


y2 (x) =
( 2 )y(x, )

=2

(2.23)

The Frobenius solution can be divided into two parts:


y(x, ) =

an ()xn+ =

n=0

n=k1
X

an ()xn+ +

n=0

an ()xn+

n=k

so that the coefficient ak is the first term of the second series. Differentiating the
expression as given in (2.23), one obtains
"n=k1
#

X
X

[( 2 )y(x, )] =
( 2 )an ()xn+ +
( 2 )an ()xn+

n=0
n=k
= log x

n=k1
X

( 2 )an ()x

n=0

X
n=k

n+

n=k1
X

n=0

[( 2 )an ()]0 xn+ + log x

n=k

40

2 )a0n ()xn+

n=k1
X
n=0

[( 2 )an ()] xn+

an ()xn+

It should be noted that an () = gn ()/f ( + n) does not contain the term ( 2 )


in its denominator until n = k, thus
( 2 )an ()|=2 = 0

for n = 0, 1, 2, . . . , k 1

( 2 )a0n ()|=2 = 0
Therefore, the second solution takes the form:

[( 2 )y(x, )]=2

n=k1
X
X
X
0
n+2
n+2
+ log x
+
[( 2 )an ()]=2 xn+2
=
an (2 )x
[( 2 )an ()]=2 x

y2 (x) =

n=0

n=k

n=k

(2.24)
It can be shown that the last infinite series is proportional to y1 (x).

41

Example: Two roots are distinct and do not differ by an integer


Obtain the power series solutions of the following differential equation about
x0 = 0:


2
1
dy
2
2d y
y=0
x
+x + x
dx2
dx
9
Since x = 0 is an RSP, use a Frobenius solution about x0 = 0:
y=

an xn+

n=0

Substituting in the differential equation results in:





X
X
1
2
n+2
an xn+ = 0
(n + )
an x
+
9
n=0
n=0
Extracting the first two lowest powered terms of the first series, such that each of the
remaining series starts with x , one obtains:




1
1
2
2
2
x
a0 x
+ ( + 1)
a1 x1
9
9



X
X
1
2
n+2
+
(n + )
an xn+ = 0
an x
+
9
n=2
n=2
Changing the indices n to m + 2 in the first series and to m in the second and
combining the two resulting series:




1
1
2
2
2

a0 x
+ ( + 1)
a1 x1
9
9



X
1
2
am+2 + am xm+ = 0
+
(m + + 2)
9
m=0
Equating the coefficients of x1 and xm+ to zero and assuming a0 6= 0 results
in the following formulas:


1
2
a0 = 0

9


1
( + 1)2
a1 = 0
9
am
am+2 =
m = 0, 1, 2, 3, . . .
(m + + 2)2 19
The characteristic equation becomes


1

9
2

42


=0

The two roots are 1 = 1/3 and 2 = 1/3. Note that 1 6= 2 and 1 2 = 2/3 is
not an integer.
Since = 1/3, then ( + 1)2

1
9

6= 0 so that the odd coefficients vanish:

a1 = a3 = a5 = . . . = 0
and
am+2 =

am
(m + + 5/3) (m + + 7/3)

with
a0
( + 5/3) ( + 7/3)
a2
a0
a4 () =
=+
( + 11/3) ( + 13/3)
( + 5/3) ( + 7/3) ( + 11/3) ( + 13/3)
a2 () =

and by induction
a2m () =

(1)m a0


7
( + 5/3) ( + 7/3) + 6m1

+
+
3
3

13
3

These coefficients are substituted in the Frobenius series:


y(x, ) =

a2m ()x2m+

m=0

For the first solution corresponding to the larger root 1 = 1/3:


 

X
1
1/3
a2m
y1 (x) = a0 x +
x2m+
3
m=1
where

 
1
= (1)m m
a2m
3
2 m!


2 m
3

a0
4 7 10 (3m + 1)

Letting = 2 = 1/3 gives the second solution:


 

X
1
1/3
y2 (x) = a0 x
+
a2m
x2m1/3
3
m=1
where


a2m

= (1)m

2m m!


2 m
3

a0
2 5 8 (3m 1)

The final solution y(x), setting a0 = 1 in each series gives:


y(x) = c1 y1 (x) + c2 y2 (x)

43

6m+1
3

Example: Two identical roots 1 = 2 = 0


Solve the following differential equation about x0 = 0:
x2

d2 y
dy
3x + (4 x) y = 0
2
dx
dx

Since x = 0 is an RSP, use a Frobenius solution about x0 = 0:


y=

an xn+

n=0

Substituting in the differential equation results in:

(n + 2) an x

n+2

an xn+1 = 0

n=0

n=0

Removing the first term and substituting n = m + 1 in the first series and n = m in
the second series results in the following equation:
2

( 2) a0 x

X


+
(m + 1)2 am+1 am xm+1 = 0
m=0

Equating the coefficients of x2 to zero, one obtains with a0 6= 0:


( 2)2 = 0

or

1 = 2 = 2 = 0

Equating the coefficient of xm+1 to zero, one obtains the recurrence formula in the
form:
am
am+1 =
m = 0, 1, 2, . . .
(m + 1)2
where
a1 =

a0
( 1)2

a2 =

a1
a0
=
2

( 1)2 2

and by induction
an () =

1)2 2 (

a0
+ 1)2 ( + n 2)2

Thus, the first solution corresponding to = 0 becomes:


y1 (x) = y(x, )|=0 =2 = a0 x2 +

X
n=1

where 0! = 1 and a0 was set to be 1, i.e., a0 = 1.

44

X xn+2
a0
n+2
x
=
12 22 n2
(n!)2
n=1

To obtain the second solution in the form, one needs a0n ():


dan ()
2a0
1
1
1
1
0
an () =
=
+ +
+ +
d
( 1)2 2 ( + 1)2 ( + n 2)2 1 + 1
+n2
Thus
a0n ()|=0 =2



2a0
1 1 1
1
= 2 2
+ + + +
1 2 n2 1 2 3
n

Defining g(n) as
g(n) =

1
1 1 1
+ + + + ,
1 2 3
n

then
a0n (0 ) =

2a0
g(n),
(n!)2

with

g(0) = 0

n = 1, 2, 3, . . .

Thus, setting a0 = 1, the second solution of the differential equation takes the form:

X
x2
g(n)
y2 (x) = y1 (x) log x 2
(n!)2
n=1

45

Example: Distinct roots that differ by an integer 1 2 = integer


Solve the following differential equation about x0 = 0:


2
9
dy
2d y
2
x
y=0
+x + x
dx2
dx
4
Since x = 0 is an RSP, use a Frobenius solution about x0 = 0:
y=

an xn+

n=0

Substituting in the differential equation results in:





X
X
9
2
(n + )
an xn+2
an xn+ = 0
4
n=0
n=0
which, upon extracting the two terms with the lowest powers of x, gives:







X
9
9
9
2
2
2
2
1

a0 x + ( 2)
a1 x +
(m + + 2)
am+2 + am xm+ = 0
4
4
4
m=0
Thus, equating the coefficient of each power of x to zero, one obtains:


9
2
a0 = 0

4


9
2
a1 = 0
( 2)
4
and the recurrence formula:
am+2 =

am
(m + + 2)2

9
4

am
(m + + 1/2) (m + + 7/2)

m = 0, 1, 2, . . .

Solving for the roots of the characteristic equation gives:


1 =

3
2

2 =

3
2

1 2 = 3 = k

Using the recurrence formula to evaluate higher ordered coefficients, one obtains:
a0
( + 1/2) ( + 7/2)
a1
=
( + 3/2) ( + 9/2)
a2
a0
=
=
( + 5/2) ( + 11/2)
( + 1/2) ( + 5/2) ( + 7/2) ( + 11/2)
a3
a1
=
=
( + 7/2) ( + 13/2)
( + 3/2) ( + 7/2) ( + 9/2) ( + 13/2)

a2 =
a3
a4
a5

46

Thus, the odd and even coefficients an can be written in terms of a0 and a1 by
induction as follows:
a0
a2m = (1)m
( + 1/2) ( + 5/2) ( + 2m 3/2) ( + 7/2) ( + 11/2) ( + 2m + 3/2)
a0
a2m+1 = (1)m
( + 3/2) ( + 7/2) ( + 2m 1/2) ( + 9/2) ( + 13/2) ( + 2m + 5/2)
To obtain the first solution corresponding to the larger root 1 = 3/2:
a0 = indeterminate
a1 = a3 = a5 = a7 = = 0
3 a0 (2m + 2)
a2m (3/2) = (1)m
(2m + 3)!

m = 1, 2, 3,

To obtain the first solution corresponding to the smaller root 2 = 3/2:


3
2 = ,
2
a0 = indeterminate

where

2 2 = 3

a1 = 0
a2m (3/2) = (1)m

a0 (2m 1)
(2m)!

m = 1, 2, 3,

The coefficient ak = a3 must be calculated to decide whether to use the second form
of the solution. Using the recurrence formula for 2 = 3/2 gives:
0
= indeterminate
0
So that the coefficient a3 is not unbounded and can be used to start a new series:


(1)m+1 a3

a2m+1 =

( + 7/2) ( + 2m 1/2) ( + 13/2) ( + 2m + 5/2)
a3 =

2 ==3/2

6 a3 m
= (1)m+1
(2m + 1)!

m = 2, 3, 4,

Thus, the second solution is obtained in the form:

2m3/2
X
X
m x2m3/2
3/2
m (2m 1)x
3/2
y2 (x) = a0 x
a0
(1)
+ a3 x + 6a3
(1)m+1
(2m)!
(2m + 1)!
m=0
m=2
= a0

m (2m

(1)

m=0

X
1)x2m3/2
(m + 1) x2m+3/2
6a3
(1)m
(2m)!
(2m + 3)!
m=0

Note that the solution starting with ak = a3 is y1 (x), which is extraneous. Letting
a0 = 1 and a3 = 0, the second solution becomes:

X
(2m 1) x2m3/2
y2 (x) =
(1)m
(2m)!
m=0

47

Example: Distinct roots that differ by an integer 1 2 = integer


Solve the following differential equation about x0 = 0:
x2

d2 y
(x + 2) y = 0
dx2

Since x = 0 is an RSP, use a Frobenius solution about x0 = 0:


y=

an xn+

n=0

Substituting the Frobenius solution in the differential equation gives:


( 2) ( + 1) a0 x

[(m + 2) (m + + 2) am+2 am ] xm+ = 0

m=0

Equating the two terms to zero gives the characteristic equation:


( 2) ( + 1) a0 = 0
Solving for the roots of the characteristic equation gives:
1 = 2

2 = 1

1 2 = 3 = k

And the recurrence formula becomes:


am+2 =

am
(m + 2) (m + + 2)

m = 0, 1, 2, . . .

Using the recurrence formula, one obtains:


a0
( 1) ( + 2)
a1
a0
a2 =
=
( + 3)
( 1) ( + 2) ( + 3)
a2
a0
a3 =
=
( + 1) ( + 4)
( 1) ( + 1) ( + 2) ( + 3) ( + 4)
a1 =

and the induction:


an () =

a0
( 1) ( + n 2) ( + 2) ( + 3) ( + n + 1)

The solution corresponding to the larger root 1 = 2:


an (2) =

6a0
n!(n + 3)!

48

n = 0, 1, 2, . . .

so that the first solution corresponding to the larger root is:


y1 (x) =

X
n=0

xn+2
n!(n + 3)!

where 6a0 was set to 1.


The solution corresponding to the smaller root 2 = 1 can be obtained after
checking a3 (1):
a3 (1)
Using the expression for the second solution in (*****), one obtains:
y2 (x) =

2
X

an (1)x

n1

n=0

[( + 1) an ()]0=1 xn1

n=3

+ log x

[( + 1) an ()]=1 xn1

n=3

Substituting for an () and performing differentiation with results in:


a0
( 1) ( + n 2) ( + 2) ( + 3) ( + n + 1)
a0
a0
( + 1) an ()|=1 =
=
( 1) ( + n 2) ( + 2) ( + 3) ( + n + 1)
2(n 3)!n!


a0
0
[( + 1) an ()] =
( 1) ( + 2) ( + n 2) ( + 2) ( + 3) ( + n + 1)


1
1
1
1
1
1
1
+ +
+ +
+
+
+ +
+

1 +2
n 2 + 2 + 3
+n+1


a0
0
[( + 1) an ()] |=1 =
(2) (2) 1 2 . . . (n 3) 1 2 . . . n


1
1
1
1
1
1+1+
+ +
+ 1 + + + +
2
+2
n 3
2
n


a0
3
=
+ g(n 3) + g(n)
2(n 3)!n!
2
( + 1) an () =

where
g(n) = 1 +

1 1
1
+ + +
2 3
n

g(0) = 0

The second solution can thus be written in the form:





1 x 1 X xn1
3
1
y2 (x) = x +
+ g(n 3) + g(n)
2 4 2 n=3 (n 3)!n!
2

X
1
xn1
+ log x
2
(n 3)!n!
n=3

49

which, upon shifting the indices in the infinite series gives:


y2 (x) = x




1 x 1 X xn+2
3
+
+ g(n) + g(n + 3)
2 4 2 n=0 n!(n + 3)!
2

X
1
xn+2
+ log x
2
(n + 3)!n!
n=0

The first series can be shown to be 43 y1 (x) which can be deleted from the second
solution, resulting in a final form for y2 (x) as:

y2 (x) = x1

1
1 x 1 X xn+2
+
[g(n) + g(n + 3)] + log x y1 (x)
2 4 2 n=0 n!(n + 3)!
2

50

Chapter 3
Special Functions
3.1

Bessel Functions

Bessel functions ares solutions to the second order differential equation:


2
2d y
x
dx2

+x


dy
+ x2 p 2 y = 0
dx

(3.1)

where x = 0 is a RSP(regular singular point) and p is a real constant.


Substituting a Frobenius solution
y(x) =

an xn+

n=0

into (3.1) results in the series:


2

a0 x

X





2
2
1
+ ( + 1) p a1 x +
(m + 2 + )2 p2 am+2 + am = 0
m=0

Equating each term to zero gives:



2 p 2 a0 = 0


( + 1)2 p2 a1 = 0
am+2 =

am
(m + 2 + )2 p2

For a0 6= 0, the characteristic equation is obtained as



2 p2 = 0
51

and its solutions are


2 = p,

1 = p,

1 2 = 2p :

The solution corresponding to the larger root 1 = p can be obtained first.


Excluding the case of p = 21 , then
a1 = a3 = a5 = = 0
am
m = 0, 1, 2, . . .
am+2 =
(m + 2)(m + 2 + 2p)
a0
a2 = 2
2 1!(p + 1)
a2
a0
a4 =
= 4
4(2 + 2p)
2 2!(p + 1)(p + 2)
a2
a0
a6 =
= 6
6(6 + 2p)
2 3!(p + 1)(p + 2)(p + 3)
and, by induction
a2m = (1)m

22m m!(p

a0
+ 1)(p + 2)(p + 3) (p + m)

m = 1, 2, 3, . . .

(3.2)

Thus, the solution corresponding to 1 = +p becomes


p

y1 (x) = a0 x + a0

x2m+p
22m m!(p + 1)(p + 2)(p + 3) (p + m)

(1)m

m=1

Using the definition of Gamma function,


Z
(x) =
tx1 et dt
0

(x + 1) = x(x)
(n + 1) = n!

where n = positive integer

then one can rewrite the expression for y1 (x) as:


y1 (x) = a0 xp + a0

(1)m

m=1

"
= a0 (p + 1)2p

(p + 1)x2m+p
22m m!(p + m + 1)

X
(x/2)p
(x/2)2m+p
+
(1)m
(p + 1) m=1
m!(p + m + 1)

Define the bracketed series as:


Jp (x) =

(1)m

m=0

52

(x/2)2m+p
m!(p + m + 1)

(3.3)

where a0 (p + 1)2p was set equal to 1 in y1 (x). The solution Jp (x) in (3.3) is known
as the Bessel function of the first kind of order p.
The solution corresponding to the smaller root 2 = p can be obtained by
substituting p into +p in (3.3) resulting in:

(x/2)2mp
y2 (x) = Jp (x) =
(1)
m!(p + m + 1)
m=0
m

(3.4)

Jp (x) in (3.4) is known as the Bessel function of the second kind of order p.
If p 6= integer, then the solution to the differential equation in (3.1) can be
written as
yh (x) = c1 Jp (x) + c2 Jp (x)
The expression for the Wronskian can be obtained form the form given in (1.4)
and (1.8):
 Z x 
d
W0
W (x) = W0 exp
= W0 e log x =

x
W0
W [Jp (x), Jp (x)] = Jp (x)Jp0 (x) Jp0 (x)Jp (x) =
x
Thus,
lim xW (x) = W0
x0

To calculate W0 , it is necessary for the leading terms only. Thus,


2p
2
lim xW [Jp (x), Jp (x)] = W0 =
=
x0
(p + 1)(1 p)
(p)(1 p)
But,
(p)(1 p) =

sin p

Then, the Wronskian is given by


W [Jp (x), Jp (x)] =

2 sin p
x

(3.5)

Another solution form first introduced by Weber, takes the form:


Yp (x) =

cos p Jp (x) Jp (x)


sin p

p 6= integer

Then, the general solution can be written in the form known as Weber function:
yh (x) = c1 Jp (x) + c2 Yp (x)

p 6= integer

The Wronskian of Weber function is obtained as


W [Jp (x), Yp (x)] = Jp (x)Yp0 (x) Yp0 (x)Jp (x) =

53

2
x

(3.6)

3.2

Bessel Function of Order Zero

If p = 0, then 1 = 2 = 0(repeated roots), which results in a solution of the form:


J0 (x) =

(1)m

m=0

X
(x/2)2m
(x/2)2m
=
(1)m
m!(m + 1) m=0
(m!)2

(3.7)

To obtain the second solution, the method developed in Section (2.4) are applied.
From the recurrence formula in (3.2), one obtains the following by setting p = 0:
am+2 =

am
(m + + 2)2

m = 0, 1, 2, . . .

Again, by induction, one can show that the even indexed coefficients are:
am+2 = (1)m

( +

2)2 (

and

y(x, ) = a0 x + a0

a0
+ 4)2 ( + 2m)2

(1)m

m=1

m = 0, 1, 2, . . .

x2m+
( + 2)2 ( + 4)2 ( + 2m)2

Using the form for the second solution given in (2.20), one obtains:

X
y(x, )
(1)m x2m+
= a0 x log x + a0 log x

( + 2)2 ( + 4)2 ( + 2m)2


0 =0
m=1



X
1
1
1
x2m+

m
+
+ +
2a0
(1)

2
2
2
( + 2) ( + 4) ( + 2m) + 2 + 4
+ 2m 0 =0
m=1

y2 (x) =

which results in the second solution y2 (x) as:


y2 (x) = log xJ0 (x) +

(1)m+1

m=0

where g(m) = 1 + 12 + 13 + +

(x/2)2m
g(m)
(m!)2

1
.
m

Define
2
[y2 (x) + ( log 2)J0 (x)]
(
)

2m
X
(x/2)
2
=
[log(x/2) + ] J0 (x) +
(1)m+1
g(m)
2

(m!)
m=0

Y0 (x) =

where is the Euler Constant,


= lim [g(n) = log n] = 0.5772 . . .
n

54

(3.8)

Since Y0 (x) is a linear combination of J0 (x) and y2 (x), it is also a solution of


(3.1). Y0 (x) is known as the Bessel function of the second kind of order zero
or the Neumann function of order zero.
Thus, the solution to (3.1) when p = 0 is:
yh (x) = c1 J0 (x) + c2 Y0 (x)

3.3

p=0

Bessel Function of an Integer Order n

If p = n = integer 6= 0, then 1 2 = 2n is an even number integer.


The solution corresponding to 1 = +n can be obtained from (3.3) by substituting p = n, resulting in:

X
(x/2)2m+
(1)m
(3.9)
Jn (x) =
m!(m
+
n)!
m=0
To obtain the second solution for 2 = n, it is necessary to check a2n (n) for
boundedness. Substituting p = n in the recurrence formula (3.2) gives:
am
am+2 =
m = 0, 1, 2, . . .
(m + 2 + n)(m + 2 + + n)
and
a1 = a3 = a5 = = 0
so that the even indexed coefficients are given by:
am+2 =

(1)m a0
( + 2 n) ( + 2m n)( + 2 + n) ( + 2m + n)

It is seen that the coefficient a2n (n) becomes unbounded, so that the methods of
the solution outlined in Section (2.4) must now be followed:
y(x, ) = a0 x +a0

(1)m

m=1

x2m+
( + 2 n) ( + 2m n)( + 2 + n) ( + 2m + n)

Then, the second solution for the case of an integer difference k = 2n results:
y2 (x) = a0

n1
X

(1)m

m=1

x2mn
(2 2n)(4 2n) (2m 2n) 2 4 (2m)


X

0
(1)m ( + n)
+ a0
x2mn
(
+
2

n)

(
+
2m

n)(
+
2
+
n)

(
+
2m
+
n)
=n
m=n



X
(1)m ( + n)
+a0 log x
x2mn
(
+
2

n)

(
+
2m

n)(
+
2
+
n)

(
+
2m
+
n)
=n
m=n
55

Thus, the solution corresponding to the second root2 = n becomes


y2 (x) =

n1
1 X (x/2)2mn
(n m 1)! + log xJn (x)
2 m=1
m!

1
1 X (x/2)2mn
+ g(n 1)Jn (x)
[g(m) + g(m + n)]
2
2 m=0 m!(n m 1)!

n+1

02
where a(n1)!
was set equal to one.

The second solution includes the first solution given in (3.9) multiplied by
1), which is a superfluous part of the second solution. Thus, removing this
component results in an expression for the second solution:
1
g(n
2

n1
1 X (x/2)2mn
(n m 1)!
y2 (x) = log xJn (x)
2 m=1
m!

1 X (x/2)2mn
[g(m) + g(m + n)]
2 m=0 m!(n m 1)!

Define
2
[( log 2) Jn (x) + y2 (x)]
(
n1
2
1 X (x/2)2mn
=
[ + log(x/2)] Jn (x)
(n m 1)!

2 m=1
m!

Yn (x) =

1 X (x/2)2mn

[g(m) + g(m + n)]


2 m=0 m!(n m 1)!

(3.10)
where Yn (x) is known as the Bessel function of the second kind of order n, or
the Neumann function of order n.
Thus, the solutions for p = n is:
yh (x) = c1 Jn (x) + c2 Yn (x)

p = n = integer

The solution of (3.1) are also known as Cylindrical Bessel functions.


The second solution of (3.10) corresponds to that given by Weber for non-integer
orders defined in (3.1).
As p n
sin p 0
cos p (1)n
56

and
Jn (x) = (1)n Jn (x)
then the form (3.1) results in an indeterminate function. Thus,
cos pJp (x) Jp (x)
pn
sin p


p (x)
sin pJn (x) + cos p Jp

Yn (x) = lim
=

cos p

Jp (x)

p

p=n


1
n
Jp (x) (1)
Jp (x)
=
p
p
p=n

(3.11)

It can be shown that this solution is a solution to (3.11). The expression in (3.11)
gives the same expression given by (3.10).

3.4

Recurrence Relations for Bessel Functions

Recurrence relations between Bessel functions of various orders are of importance


because of their use in numerical computations of high order Bessel functions.
Starting with the definition of Jp (x) in (3.3),

(x/2)2m+p
Jp (x) =
(1)
m!(p + m + 1)
m=0
m

then differentiating the expression in (3.3), one obtains:


Jp0 (x)

x 2m+p
1X
m [2(m + p) p] 2
=
(1)
2 m=0
m!(p + m + 1)




x 2m+p1
x 2m+p x 1
X
pX
m [2(m + p) p] 2
m
2
2
=
(1)

(1)
m!(p
+
m
+
1)
2
m!(p + m + 1)
m=0
m=0

Using (p + m + 1) = (m + p)(m + p), then


p
Jp0 (x) = Jp1 (x) Jp (x)
x
Another form of (3.12) can be obtained. Again, starting with Jp0 (x)
Jp0 (x)

x 2m+p1
pX
m
2
=
(1)
+
(1)
(m 1)!(p + m + 1) 2 m=0
m!(p + m + 1)
m=0


x 2m+p1
2

57

(3.12)

Since (m 1)! for m = 0, then


Jp0 (x)


x 2m+p1
2

p
+ Jp (x)
m!(p + m + 1) x
m=1


x 2m+p+1
X
p
m+1
2
=
(1)
+ Jp (x)
m!(p + m + 2) x
m=0
=

(1)

Hence,

p
(3.13)
Jp0 (x) = Jp+1 (x) + Jp (x)
x
Combining (3.12) and (3.13), one obtains another expression for the derivatives:
Jp0 (x) = [Jp1 (x) Jp+1 (x)]

(3.14)

Equating (3.12) and (3.13), one obtains a recurrence formula for Bessel functions of
order (p + 1) in terms of orders p and p 1:
Jp+1 (x) =

2p
Jp (x) Jp1 (x)
x

(3.15)

Multiplying (3.13) by xp , and rearranging the resulting expression, one obtains:


1 d  p
x Jp+ (x) = x(p+1) Jp+1 (x)
x dx

(3.16)

If p is substituted by p + 1 in the form given in (3.16) this results in:


1 d  (p+1)
x
Jp+1 = x(p+2) Jp+2
x dx

then upon substitution of (3.16), one obtains


2

 p 
1 d
2
(1)
x Jp = x(p+2) Jp+2
x dx
Thus, by induction, one obtains a recurrence formula for Bessel Functions:

r
 p 
1 d
r
(1)
x Jp = x(p+r) Jp+r
r0
x dx
Substitution of p by p in (3.17) results in another recurrence formula:

r
1 d
r
[xp Jp ] = xpr Jp+r
r0
(1)
x dx

(3.17)

(3.18)

Substitution of p by p in (3.12), one obtains:


0
Jp
px1 Jp = J(p+1)

58

(3.19)

Multiplying (3.19) by xp , one obtains a new recurrence formula:



1 d  p
x Jp (x) = x(p+1) J(p+1) (x)
x dx

(3.20)

Substitution of p + 1 for p in (3.19) results in the following equation:



1 d  (p+1)
x
J(p+1) (x) = x(p+2) J(p+2) (x)
x dx
or upon substitution of (3.20) one gets:


1 d
x dx

2


xp Jp = x(p+2) J(p+2)

and, by induction, a recurrence formula for negative ordered Bessel functions is obtained:

r
 p

1 d
x Jp = x(p+r) J(p+r)
r0
(3.21)
x dx
Substitution of p by p in (3.21) results in the following equation:
r

1 d
[xp Jp ] = xpr Jpr
r0
x dx

(3.22)

To obtain the recurrence relationships for Yp (x), it is sufficient to use the form
of Yp (x) given in (3.1) and the recurrence equations given in (3.17), (3.18), (3.21),
and (3.22).
Starting with (3.17) and (3.21) and setting r = 1, one obtains
1 d  p 
x Jp = x(p+1) Jp+1
x dx

1 d  p
x Jp = x(p+1) J(p+1)
x dx
Then, using the form in (3.1) for Yp (x):



 1 d
1 d  p
cos p Jp Jp (x)
p
x Yp =
x
x dx
x dx
sin p


(p+1) cos(p + 1) Jp+1 J(p+1) (x)
= x
= x(p+1) Yp+1
sin(p + 1)
such that:
x Yp0 p Yp = x Yp+1
Similarly, use of (3.18) and (3.22)
x Yp0 + p Yp = x Yp1
59

Combining the above two formulas, the following recurrence formulas can be derived:
2p
Yp
x
= 2 Yp0

Yp1 + Yp+1 =
Yp1 Yp+1

The recurrence relationships developed for Yp are also valid for integer values of p,
since Yn can be obtained from Yp by the expression given in (3.11).
The recurrence formulas developed in this section can be summarized as follows:
p
Z0p = Zp+1 + Zp
x
p
0
Zp = Zp1 Zp
x
1
Z0p =
(Zp1 Zp+1 )
2
2p
Z0p+1 = Zp1 + Zp
x

(3.23)
(3.24)
(3.25)
(3.26)

where Zp denotes Jp (x), Jp (x), or Yp (x) for all values of p.

3.5

Bessel Functions of Half Orders

If the parameter p in the Bessel differential equation happens to be an odd multiple


of 21 ,

dy
d2 y
x2 2 + x + x2 p 2 y = 0
dx
dx
then it is possible to obtain a closed form of Bessel functions of half orders.
Starting with the lowest half order, p = 1/2, then using the form in (3.3), one
obtains:


1 2m+1/2
X
2
J1/2 (x) =
(1)m
m!
(m + 3/2)
m=0

 x 1/2 X
x2m
(1)m m m
=
2
2 (2 m!) (m + 3/2)
m=0
which can be shown to result in the following closed form:

J1/2 (x) =

2
x

1/2 X

x2m+1
(1)
=
(2m + 1)!
m=0
m

60

2
x

1/2
sin x

(3.27)

Similarly, it can be shown that



J1/2 (x) =

2
x

1/2
cos x

(3.28)

To obtain the higher ordered half-ordered Bessel functions Jn+1/2 and J(n+1/2) ,
one can use the recurrence formulas in (3.23)(3.26). One can also obtain these
expressions by using (3.17) and (3.21) by setting p = 1/2, resulting in the following
expressions:

n 

p
1 d
sin x
n
n+1/2
(3.29)
Jn+1/2 = (1) 2/x
x dx
x
n 

p
cos x 
1 d
n+1/2
J(n+1/2) =
2/x
(3.30)
x dx
x

3.6

Hankel Functions

Hankel functions are complex linear combinations of Bessel functions of the form:
Hp(1) = Jp + i Yp

(3.31)

Hp(2) = Jp i Yp

(3.32)

(1)
(2)
where i = 1. Hp and Hp in (3.31) and (3.32) are respectively known as the
Hankel functions of first and second kind of order p.
The general solution of (3.1) can be written in the form:
y(x) = c1 Hp(1) + c2 Hp(2)
Recurrence formulas for Hankel functions take the same forms given in (3.23)(3.26),
since they are linear combinations of Jp and Yp .

61

3.7

Modified Bessel Functions

Modified Bessel functions are solutions to a differential equation different from that
given in (3.1). Especially, they are solutions to the following differential equation:
x2


d2 y
dy
2
2
y=0
+
x
+
p
+
x
dx2
dx

(3.33)

Performing the transformation:


z = ix
then the differential equation in (3.33) transforms to:
z2


d2 y
dy
2
2
+
z
+
z

p
y=0
dz 2
dz

which has two solutions of the form given in (3.3) and (3.4) if p 6= 0 and p 6= integer.
Use the form in (3.3), one obtains

Jp (z) =

(1)m

m=0

(z/2)2m+p
m!(p + m + 1)

p 6= 0, 1, 2, 3,

X
(ix/2)2m+p
(ix/2)2m+p
p
Jp (ix) =
(1)
= (i)
m!(p + m + 1)
m!(p + m + 1)
m=0
m=1
m

and

X
(x/2)2mp
(ix/2)2mp
p
= (i)
Jp (ix) =
(1)
m!(p + m + 1)
m!(p + m + 1)
m=0
m=0
m

Define:
Ip (x) =

(x/2)2m+p
= (i)p Jp (ix)
m!(m
+
p
+
1)
m=0

(x/2)2mp
Ip (x) =
= (i)p Jp (ix)
m!(m

p
+
1)
m=0

(3.34)
p 6= 0, 1, 2, 3, (3.35)

Ip (x) and Ip (x) are known, respectively, as the modified Bessel function of the
first and second kind of order p.
The general solution of (3.33) takes the following form:
y(x) = c1 Ip (x) + c2 Ip (x)

62

If p takes the value zero or an integer n, then




x 2m+n
X
2
n = 0, 1, 2, 3,
In (x) =
m!(m + n)!
m=0

(3.36)

The second solution must be obtained in a similar manner as in Sections (3.2) and
(3.3) giving
n+1

Kn (x) = (1)

n1
h x
i
1X
(n m 1)!  x 2mn
log
+ In (x) +
(1)m
2
2 m=0
m!
2

x 2mn

(1)n X
2
[g(m) + g(m + n)] ,
+
2 m=0 m!(m + n)!

n = 0, 1, 2, 3, (3.37)

The second solution can also be obtained from a definition given by Macdonald:


Ip Ip
Kp =
(3.38)
2 sin(p)
Kp is known as the Macdonald function. If p is an integer equal to n, then taking the
limit p n:


1
Ip
n Ip
Kn = (1)
(3.39)

2
p
p p=n
If p is 1/2, then the modified Bessel functions of half-orders can be developed in
a similar manner, resulting in
r
2
sinh x
I1/2 =
x
r
2
I1/2 =
cosh x
x

63

3.8

Generalized Equations Leading to Solutions in


Terms of Bessel Functions

The differential equation in (3.1) leads to solutions Zp (x), with Zp (x) representing Jp ,
(1)
(2)
Yp , Jp , Hp , Hp . One can obtain the solutions of different and more complicated
equations in terms of Bessel functions.
Starting with an equation of the form:
x2


d2 y
dy
2 2
2
y=0
+
k
x

r
+
(1

2a)
dx2
dx

(3.40)

a solution of the form:


y = x u(x)
can be tried, resulting in the following differential equation:
x2


d2 u
du 
+ x + k 2 x 2 r 2 + a2 u = 0
2
dx
dx

where was set equal to a.


Furthermore, if one lets z = kx, then
z2

d
dx


du
d2 u
2
2
+
z
+
z

p
u=0
dz 2
dz

d
= k dz
and

with p2 = r2 + a2

whose solution becomes:


u = c1 Jp (z) + c2 Yp (z)
Thus, the solution to (3.40) becomes:
y(x) = xa [c1 Jp (kx) + c2 Yp (kx)]

(3.41)

where p2 = r2 + a2 .
A more complicated equation can be developed form from (3.40) by assuming
that


dy
d2 y
+ (1 2a)z + z 2 r2 y = 0
2
dz
dz
which has solutions of the form:
z2

u = z a [c1 Jp (z) + c2 Yp (z)]


where p2 = r2 + a2 .

64

(3.42)

(3.43)

If one lets z = f (x), then (3.42) transforms to:





f 0 f 00 dy (f 0 )2 2
d2 y
+ 2 f r2 y = 0
+ (1 2a) 0
2
dx
f
f dx
f

(3.44)

whose solutions can be written as


y(x) = [f (x)]a {c1 Jp [f (x)] + c2 Yp [f (x)]}
where p2 = r2 + a2 .
Equation (3.44) may have many solutions depending on the desired form of f (x):
(i) If f (x) = k xb , then the differential equation may be written as
x2


dy
d2 y
2
2 2b
2
+
(1

2ab)x
+
b
k
x

r
y=0
dx2
dx

(3.45)

whose solution is given by





y(x) = xab c1 Jp kxb ) + c2 Yp kxb )

(3.46)

(ii) If f (x) = k ebx , then the differential equation may be written as



d2 y
dy
2ab
+ b2 k 2 e2bx r2 y = 0
2
dx
dx

(3.47)

whose solution is given by





y(x) = xabx c1 Jp kebx ) + c2 Yp kebx )

(3.48)

Another type of a differential equation that leads to Bessel function type solutions can be obtained from the form developed in (3.44).
If one lets y to be transformed as follows: then


f 0 f 00
g 0 du
d2 u
+ (1 2a) 0 2
dx2
f
f
g dx
 0 2


 g 00 g 0
(f )
f 0 f 00
g0
2
2
+
f r

(1 2a) 0 2
u=0
f
g
g
f
f
g
whose solutions are given in the form:
u(x) = g(x) [f (x)]a {c1 Jp [f (x)] + c2 Yp [f (x)]}

65

(3.49)

where p2 = r2 + a2 . If one lets:


g(x) = ecx
f (x) = kxb
then, the differential equation has the form:
x2



d2 u
du  2 2 2b
2 2
2
+
c
x

cx
(1

2ab)
u = 0 (3.50)
k
x

r
+
b
+
[1

2ab

2cx)
x
dx2
dx

whose solutions are expressed in the form:





u(x) = ecx xab c1 Jp kxb + c2 Yp kxb

66

(3.51)

3.9

Zeroes(Roots) of Bessel Functions

Bessel functions Jp (x) and Yp (x) have infinite number of zeroes, which are the roots
of the Bessel functions. That is, x is a roots of Jp (x) = 0 or Yp (x) = 0.
0
0
Denoting the sth root of Jp (x), Yp (x), Jp0 (x), and Yp0 (x) by jp,s , yp,s , jp,s
, and yp,s
,
then all the zeroes of these functions have have the following properties:

1. All zeroes of these Bessel functions are real if p is real and positive.
2. There are no repeated roots, except at the origin.
3. jp,0 = 0 for p = 0.
4. The roots of Jp (x) and Yp (x) interlace, such that
p < jp,1 < jp+1,1 < jp,2 < jp+1,2 < jp,3 <
p < yp,1 < yp+1,1 < yp,2 < yp+1,2 < yp,3 <
0
0
0
0
0
0
p jp,1
< yp,1
< jp,2
< yp,2
< jp,3
< yp,3
<

5. The roots can be bracketed such that


p
p
p(p + 2) < jp,1 < 2(p + 1)(p + 3)
p
p
0
p(p + 2) < jp,1
< 2(p + 1)(p + 1)
6. The large roots of Bessel functions for a fixed order p take the following asymptotic
form:


p 1
As s , jp,s l +

2 4


p 3
As s , yp,s l +

2 4


p 3
0
As s , jp,s l +

2 4


p 1
0
As s , yp,s l +

2 4

The large zeros of the spherical Bessel functions of order n are the same as the zeroes
of Jp (x), Yp (x), Jp0 (x), and Yp0 (x).
67

3.10

Graphs of Bessel Functions

J0 (x)
1.0
0.8
0.6
0.4
0.2

10

15

20

25

30

-0.2
-0.4

Figure 3.1: The graph of J0 (x)

J1 (x)
0.6

0.4

0.2

10

20

30

-0.2

Figure 3.2: The graph of J1 (x)

68

40

50

J2 (x)
0.6

0.4

0.2

10

20

30

40

50

40

50

-0.2

Figure 3.3: The graph of J2 (x)

J4.5 (x)

0.4

0.2

10

20

30

-0.2

Figure 3.4: The graph of J4.5 (x)

69

Comparison of J1 (x), J2 (x), and J3 (x)


0.6

0.4

0.2

10

20

30

40

-0.2

Figure 3.5: The graphs of J1 (x), J2 (x), and J3 (x)

70

50

Y0 (x)

0.4
0.2

10

15

20

25

30

25

30

-0.2
-0.4
-0.6

Figure 3.6: The graph of Y0 (x)

Y1 (x)
0.4

0.2

10

15

20

-0.2

-0.4

-0.6

Figure 3.7: The graph of Y1 (x)

71

Y2 (x)

0.2

10

15

20

25

30

25

30

-0.2

-0.4

-0.6

Figure 3.8: The graph of Y2 (x)

Y4.5 (x)
0.2

10

15

20

-0.2
-0.4
-0.6
-0.8

Figure 3.9: The graph of Y4.5 (x)

72

Comparison of Y0 (x), Y1 (x), and Y2 (x)


0.6

0.4

0.2

10

20

30

40

-0.2

Figure 3.10: The graphs of Y0 (x), Y1 (x), and Y2 (x)

73

50

Chapter 4
Boundary Value Problems and
Eigenvalue Problems
4.1

Introduction

4.1.1

Initial Value Problems

Referring to Section (1.9), Initial Value Problems(IVP) are differential equations


of order n
dn y
dn1 y
d2 y
dy
a0 (x) n + a1 (x) n1 + + an2 (x) 2 + an1 (x) + an (x)y = f (x)
dx
dx
dx
dx
together with a n conditions specified on the dependent variable and its first (n 1)
derivatives at an initial point
y(x0 ) = given
y 0 (x0 ) = given
..
.
. = ..

a x, x0 b

y (n2) (x0 ) = given


y (n1) (x0 ) = given
If the coefficients a0 (x) 6= 0, a1 (x), , an (x) of the linear ordinary differential equation are continuous and bounded in a x b, then the initial value problem has a
unique solution. The solutions to such problems are unique and valid over the range
of all values of the independent variable.

74

Example 4.1: Initial Value Problem


Obtain the solution to the following initial value problem:
Differential Equation(DE):
y 00 + 4 y = f (x) = 4x
Initial Condition(IC):
y(/4) = 2
y 0 (/4) = 3
The complete solution to the differential equation is obtained as
y(x) = c1 sin 2x + c2 cos 2x + x
The two arbitrary constants can be evaluated by applying the given initial conditions:
c1 = 2 /4

and

c2 = 1

and the unique solution is obtained as


y(x) = (2 /4) sin 2x cos 2x + x

If the differential equation is homogeneous, i.e., f (x) = 0, and the initial condition
are non-homogeneous,
Differential Equation(DE):
Initial Condition(IC):

y 00 + 4 y = f (x) = 0
y(/4) = 2
y 0 (/4) = 3

then the solution for homogeneous differential equation and homogeneous initial conditions becomes:
y(x) = 2 sin 2x

3
cos 2x
2

for all x

which is a non-trivial solution.


If the differential equation is homogeneous, i.e., f (x) = 0, and the initial condition
are homogeneous,
Differential Equation(DE):
Initial Condition(IC):

y 00 + 4 y = f (x) = 0
y(/4) = 0
y 0 (/4) = 0

then the solution vanishes identically, i.e.,


y(x) 0

for all x

which is a trivial solution.


75

4.1.2

Boundary Value Problems

Boundary Value Problems(BVP) are differential equations of order n with n


conditions specified on two end points of a bounded region valid in the closed
region between the two end points. There can be many kinds of boundary conditions.
For example, the total number of n conditions specified on two end points are given
by
y(a) = given

y(b) = given

y 0 (a) = given
..
.

y 00 (b) = given
..
.

Example 4.2: Boundary Value Problem


Obtain the solution to the following boundary value problem:
Differential Equation(DE):
y 00 + 4 y = f (x) = 4x
0 x pi/4
Boundary Condition(BC):
y(0) = 2
y(/4) = 3
The complete solution to the differential equation is obtained as
y(x) = c1 sin 2x + c2 cos 2x + x
The two arbitrary constants can be evaluated by applying the given boundary conditions:
y(0) = c2 = 2


y(/4) = c1 sin + 2 cos + = 3
2
2
4
and the unique solution is obtained as

y(x) = (3 ) sin 2x + 2 cos 2x + x


4

c1 = 3

0 x /4

If the differential equation is homogeneous, i.e., f (x) = 0, and the boundary conditions are non-homogeneous, then the solution becomes:
y(x) = 3 sin 2x + 2 cos 2x

0 x /4

If both the differential equation and the boundary conditions are homogeneous, then
the solution vanishes identically, i.e.,
y(x) 0

0 x /4

which is a trivial solution.


76

4.1.3

Eigenvalue Problems

An Eigenvalue problem is a special type of a homogeneous boundary value problem


with an undetermined parameter, which has a non-trivial solution. A non-trivial
solution exists for such problems if the parameter takes on certain values. Nontrivial solutions to an eigenvalue problem are referred as Eigenfunctions and the
corresponding undetermined parameter that takes on certain values are known as
Eigenvalues.
Example 4.3: Eigenvalue Problem
Obtain the solution to the following eigenvalue problem:
Differential Equation(DE):
y 00 + y = 0
is an unspecified parameter
Boundary Condition(BC):
y(0) = 0
y(/4) = 0
The complete solution to the differential equation becomes

y(x) = c1 sin x + c2 cos x

6= 0

y(x) = c3 x + c4

=0

Applying the boundary conditions at the two end points yields:


y(0) = c2 = 0
 
y(/4) = c1 sin
=0
4

and

c4 = 0

and

c3 = 0

The last equation on c1 leads to two possible solutions:


(i) For a non-trivial solution, i.e., c1 6= 0, then sin 4 = n , n = integers,
which can be satisfied if the undetermined parameter takes any one of the
following infinite discrete number of possible values, i.e.,:
1 = 16 12 ,

2 = 16 22 ,

3 = 16 32 ,

In other words, n = 16 n2 , n = 1, 2, 3, are the eigenvalues which satisfy


the following the characteristic equation:
 
sin
=0
4
Thus, the solution, which is nontrivial if takes any one of these special values,
has the following form:
y = c1 sin(4 n x)
77

n = 1, 2, 3,

which is non-unique, since the constant c1 is undeterminable.


The function n = sin(4 n x) are the known as Eigenfunctions. The value
= 0 gives a trivial solution, thus it is not an eigenvalue.
(ii) If does not take any one of those values, i.e., if
n 6= 16 n2 ,

n = 1, 2, 3,

then
c1 = 0
and the solution vanishes identically.

4.2

Examples of Eigenvalue Problems

These are the examples of eigenvalue problems of linear differential equations in science and engineering.
Vibration, Wave Propagation and Whirling of Stretched String
Longitudinal Vibration and Wave Propagation in Elastic Bars
Vibration, Wave Propagation and Whirling of Beams
Waves in Acoustic Horns
Stability of Compressed Columns
Torsional Vibration of Circular Bars

4.2.1

Vibration, Wave Propagation and Whirling of Stretched


String

The differential equation of motion is derived as


f (x)
d2 y
+ k2y =
2
dt
T0

78

where
y = lateral displacement

k =
= wave number
c
= angular velocity of whirling
c = the sound speed of waves

The natural(physical) boundary conditions are of three types:


1. fixed end:

y(0) = 0

or y(L) = 0

2. free end:

y 0 (0) = 0

or y 0 (L) = 0

3. elastically supported end:


- left end:

T0 dy(0)
y(0) = 0
dx

- right end: T0 dy(L)


+ y(L) = 0
dx

4.2.2

Longitudinal Vibration and Wave Propagation in Elastic Bars

One obtains the wave equation for an elastic bar:





u
2 u
EA
= A 2 f
x
x
t
if the bar is homogeneous and its cross-sectional area is constant, then the wave
equation is simplified:
2 u
1 2 u
f
= 2 2
x2
c t
AE
2
where c = E/ is the sound speed of longitudinal waves in the bar.
For a bar that is vibrating with a circular frequency , the differential equation of
motion is derived as
d2 u
f
2
+
k
u
=

dx2
AE

k=

The natural(physical) boundary conditions are of three types:


1. Fixed end:

u = 0
79

2. Free end:

AEu /x = 0

3. Elastically supported by a linear spring:


- left end:

AEu /x u = 0

- right end: AEu /x + u = 0

4.2.3

Vibration and Wave Propagation of Beams

The vibration of beams or the whirling of shafts can be considered as similar dynamic
system to the vibration or whirling of strings. The equation of motion for the beam
becomes:


d2 y
d2
EI 2 = 2 Ay + f (x)
2
dx
dx
where y is the beam displacement, is the mass density, A is the cross-sectional area,
I is the area moment of inertia.
if the EI and A are constants, the equation of motion for the beam simplifies to:
f (x)
d4 y
4

y
=
dx4
EI
where , the wave number, is defined by
A 2

EI

4 =

The natural(physical) boundary conditions are of three types:


1. fixed end:
2. free end:
3. simply supported:
4. free-fixed end:

y(0) = 0
d2 y
dx2

=0

y(0) = 0
dy
dx

=0


or
or

dy
(L)
dx
d
dx

=0

d2 y
EI dx
2

=0

d y
or EI dx
2 = 0


d2 y
d
or dx EI dx2 = 0

d2 y
EI dx
2

d y
5. elastically supported end:
y = 0
or dx
2 = 0
The + and - signs refer to the left and right ends, respectively
d
dx

80

4.3

Adjoint Systems

Consider the linear nth order differential operator L:




dn
dn1
d2
d
Ly = a0 (x) n + a1 (x) n1 + + an2 (x) 2 + an1 (x) + an (x) y = 0,
dx
dx
dx
dx

axb

(4.1)
where a0 (x) 6= 0, a1 (x), , an1 (x), an (x) are continuous and differentiable n times.
Then, define the linear nth order differential operator K:
n1
n2
dn
n1 d
n2 d
[a
(x)y]
+
(1)
[a
(x)y]
+
(1)
[a2 (x)y] +
0
1
dxn
dxn1
dxn2
d2
d
+ + (1)2 2 [an2 (x)y]
[an1 (x)y] + an (x)y (4.2)
dx
dx
as the Adjoint operator to the operator L. The differential equation:

Ky = (1)n

Ky = 0
is the adjoint differential equation of (4.1).
The operator L and its adjoint operator K satisfy the following identity:
v(x)Lu(x) u(x)Kv(x) =
where

n1 m
X
d u(x)
P (u, v) =
dxm
m=0

d
P [u(x), v(x)]
dx

(nm1
)
k
X
k d
(1)
[anmk1 (x)v(x)]
dxk
k=0

(4.3)

(4.4)

Equation (4.3) is known as Lagranges Identity.


The determinant (x) of the coefficients of the bilinear form of ui v j becomes:
(x) = [a0 (x)]n

(4.5)

which does not vanish in a x b.


Integrating (4.3), one obtains Greens formula having the form:
Z b
b

(vLu uKv) dx = P (u, v)
a

The determinant of the bilinear



(a)
0


0
(b)

form of the right side of (4.6) becomes:





n
= (a) (b) = [a0 (a)a0 (b)] 6= 0

If the operator K = L, then the operator L and K are called Self-Adjoint.


81

(4.6)

4.3.1

Self-Adjoint Systems

Self-Adjoint System for 2nd-order Differential Equations


As an example, take the general second order differential equation:
Ly = a0 (x)y 00 + a1 (x)y 0 + a2 (x)y = 0
Then, the adjoint operator K becomes:
Ky = (1)2 (a0 y)00 + (1)1 (a1 y)0 + (1)0 a2 y
= (a0 y)00 (a1 y)0 + a2 y
= a0 y 00 + (2a00 a1 ) y 0 + (a000 a01 + a2 ) y
which is not equal to Ly in general and hence the operator L is not self-adjoint. If
the operator L is self-adjoint, then the following equalities must hold:
a1 = 2a00 a1

and

a2 = a000 a01 + a2

which can be satisfied by one relationship, namely:


a00 = a1
which is not true in general. However, one can change the second-order operator L
by a suitable function multiplier so that it becomes self-adjoint, where this operation
is valid only for the second order operator. Hence, if one multipliers the operator L1
by an undetermined function z(x), then
L1 y = z(x)Ly
so that L1 is self-adjoint. Then, each coefficient is multiplied by z(x). Since the
condition for self-adjointness requires that the differential of the first coefficient of L
equals the second, then
(za0 )0 = za1
which is rewritten as

z0
a1 a00
=
z
a0

The function z(x) can be obtained readily by integrating the above differentials:
Z x

a1 ()
p(x)
1
z(x) =
exp
d =
a0 (x)
a0 ()
a0 (x)
82

Using the multiplier function z(x), the self-adjoint operator L1 can be rewritten as:
a1 (x)
a2 (x)
L1 y = p(x)y 00 +
p(x)y 0 +
p(x)y
a0 (x)
a0 (x)
 


d
d
=
p
+q y
dx
dx
where
Z
p(x) = exp

a1 ()
d
a0 ()


and

q(x) =

a2 (x)
p(x)
a0 (x)

Thus, any second order, linear differential equation can be transformed to a selfadjoint form. The method used to change a second order differential operator L
become self-adjoint cannot be duplicated for higher order equations.

Self-Adjoint System for Higher-order Differential Equations


In general, if the order n is an odd integer, then that operator cannot be self-adjoint,
since that requires a0 (x) = a0 (x). It should be noted that if the order n of the
differential operator L is an odd integer, then the differential equation is not invariant
to coordinate inversion, i.e., the operator is not the same if x is changed to (x).
Therefore, if the independent variable x is a spatial coordinate, then the operator
L, representing the systems governing equation, would have a change of sign of
the coefficient of its highest derivative if x is changed to (x). This would lead to a
solution that is drastically different from that due to an uninverted operator L. Thus,
a differential operator L which represent a physical systems governing equation on a
spatial coordinate x cannot have an odd order n.
In general, a physical system governed by a differential operator L on a spatial
coordinate is self-adjoint if the system satisfies the law of conservation of energy.
Thus, if the governing equations are derived from a Lagrange function representing
the total energy of a system, then the differential operator L is self-adjoint. A general
form of a linear, non-homogeneous (2n)th order differential operator L which is selfadjoint can be written as follows:


n
k
X
dk y
k d
Ly =
(1)
pnk (x) k
dxk
dx
k=0

n

(n1)
0
= (1)n p0 (x) y (n) + (1)n1 p0 (x) y (n1)
+ [pn1 (x) y 0 ] + pn (x) y = f (x),
(4.7)

83

4.4

Boundary Value Problems

The solution of a system is unique if and only if n conditions on the function y(x) and
its derivatives up to (n 1) are specified at the end points a and b. Thus, a general
form of non-homogeneous boundary conditions can be written as follows:
n1
X


Ui (y) =
ij y (k) (a) + ij y (k) (b) = i ,

i = 1, 2, , n

(4.8)

k=0

where ij , ij , and i are real constants. The boundary conditions in (4.8) must be
independent. This means that the determinant does not vanish:
det [ij , ij ] 6= 0
The non-homogenous differential equation (4.9) and non-homogenous boundary
conditions (4.10) constitute a general form of boundary value problems.
Ly = f (x)

(4.9)

Ui (y) = i ,

i = 1, 2, , n

(4.10)

A necessary and sufficient condition for the solution of such problems to be unique,
is that the equivalent homogenous system:
Ly = 0
Ui (y) = 0,

i = 1, 2, 3, , n

has only the trivial solution, y(x) 0. Thus, an nth -order self-adjoint operator given
in (4.1) has n independent solution yi (x).
Since the set of n homogeneous conditions given in (4.8) are independent, then
the solution of the differential equation (4.7) can be written as:
n
X
y(x) =
ci yi (x) + yp (x)
i=1

where ci are arbitrary constants. Then, there exists a non-vanishing unique set of
constants [ci ] which satisfies these boundary conditions.

Homogeneous Boundary Value Problems


A homogenous boundary value problem consists of an nth -order differential operator
and a set of n linear boundary conditions, i.e.,
Lu = 0

(4.11)

Ui (u) = 0,

i = 1, 2, , n
84

(4.12)

An adjoint system to that defined above is defined by:


Kv = 0

(4.13)

Vi (v) = 0,

i = 1, 2, 3, , n

(4.14)

where the homogeneous boundary conditions (4.14) are obtained by substituting the
boundary conditions Ui (u) = 0 in (4.8) into
b

P (u, v) = 0
(4.15)
a

with the P (u, v) given in (4.4). If the operator L is a self-adjoint operator, i.e., if
L = K, then the boundary conditions can be shown to be identical, i.e.,
Ui (u) = Vi (u)

(4.16)

Example:
Consider the operator:
Ly = a0 (x) y 00 + a1 (x) y 0 + a2 (x) y = 0

axb

the adjoint operator is given by:


My = [a0 (x) y]00 [a1 (x) y]0 + a2 (x) y = 0

axb

The bilinear form P (u, v) is given by:


b
b


P (u, v) a = u a1 v (a0 v)0 + u0 (a0 v) a = 0

(i) Consider the boundary condition pair on u given by


u(a) = 0

and

u(b) = 0

and substitution into (4.15) results in the following:


b
u0 (a0 v) a = u0 (b) [a0 (b)v(b)] u0 (a) [a0 (a)v(a)] = 0
Since u(b) = 0, then u0 (b) = 0 is an arbitrary constant. Similarly since u(b) = 0, then
u0 (a) = 0 is an arbitrary constant. For arbitrary constants u0 (a) = 0 and u0 (b) = 0,
the relation can be satisfied if:
v(a) = 0

and
85

v(b) = 0

(ii) If u0 (a) = 0 and u(b) = 0, then one obtains the following when into P (u, v) = 0:


u0 (b) [a0 (b)v(b)] u(a) a1 (a)v(a) [a0 (a)v(a)]0 = 0
Since u0 (a) = 0, then u(b) is arbitrary. Similarly, since u(b) = 0, then u0 (b) is arbitrary.
Thus, the boundary conditions Vi (v) = 0 are:
a1 (a)v(a) [a0 (a)v(a)]0 = 0

4.5

and

v(b) = 0

Eigenvalue Problems

An eigenvalue problem is a system that satisfies a differential equation with an unspecified arbitrary constant and satisfying a homogenous or non-homogeneous set
of boundary conditions.
Consider a general form of a homogenous eigenvalue problem:
Ly + My = 0

(4.17)

Ui (y) = 0,

i = 1, 2, n

where L is given by (4.1) and the boundary conditions by (4.8). The operator M is
an mth order differential operator where m < n and is an arbitrary constant.
A general form of a self-adjoint homogenous eigenvalue problem takes the following form:
Ly + My = 0

axb

Ui (y) = 0,

i = 1, 2, 2n

(4.18)

where L and M are linear self-adjoint operators of order 2n and 2m respectively,


where


n
k
X
dk y
k d
Ly =
(1)
pnk k
(4.19)
dxk
dx
k=0


m
k
X
dk y
k d
My =
(1)
qnk k
n>m
k
dx
dx
k=0
where is an undetermined parameter, and Ui (y) = 0 are 2n homogenous boundary
conditions having the same form in (4.8).

86

Comparison Function
A Comparison Function u(x) is defined as an arbitrary function that has 2n continuous derivatives and satisfies the boundary conditions Ui (u) = 0, i = 1, 2, 2n.
If u(x) and u(x) are arbitrary comparison functions, the following integrals vanish for self-adjoint eigenvalue problems:
Z b
(u Lv v Lu) d x = 0
(4.20)
a

and
Z

(u Mv v Mu) d x = 0

(4.21)

The expression for P (u, v) in (4.4) that corresponds to a differential operator L


and M given in (4.18) becomes
Z
a

b

(u Lv v Lu) d x = P (u, v)

k1
n X
X

n
o b




k+r
(r)
(k) (kr1)
(r)
(k) (kr1)
(1)
u
pnk v
v
pnk u
= 0 (4.22)
a

k=1 r=0

Similar expression for P (u, v) for the differential operator M can be developed by
substituting m and qi in (4.22) for n and pi , respectively. It is obvious that the right
side of (4.22) must vanish for the system to be self-adjoint.
An eigenvalue problem is called Positive Definite if, for every non-vanishing
comparison function u(x), the following inequalities hold:
Z

Z
u Lu dx < 0

87

u Mu dx > 0

and

(4.23)

Example:
Examine the following eigenvalue problem for self-adjointness(SA) and positivedefiniteness(PD):
y 00 + r(x)y = 0

r(x) > 0

y(a) = 0

y(b) = 0

axb

For this problem the operators L and M, defined as


L=

d2
dx2

M = r(x)

and

are self-adjoint.
Let u(x) and v(x) be comparison functions, such that
u(a) = v(a) = 0

and

u(b) = v(b) = 0

Thus, to establish if the system is self-adjoint, one substitutes into (4.20) and (4.21):
Z
a

b Z b

(u0 v 0 v 0 u0 ) dx = 0
(u v v u ) dx = (u v v u )
00

00

and

(u rv v ru) dx = 0
a

which proves that the eigenvalue problem is self-adjoint(SA).


To establish that the problem is also positive definite, substitute L and M into
(4.23):
b Z b
u u dx = u u
(u0 )2 dx < 0
a
a
a
Z b
Z b
u ru dx =
ru2 dx > 0

00

r(x) > 0

which indicates that the eigenvalue problem is also positive definite(PD).

88

4.6

Properties of Eigenfunctions of Self-Adjoint


Systems

Self-adjoint eigenvalue problems have some properties unique to this system.

(i) Orthogonal Eigenfunctions


If the eigenvalue problem is self-adjoint, then the eigenfunctions are orthogonal. Let
n and m be any two eigenfunctions corresponding to different eigenvalues n and
m , then each satisfies its respective differential equation, i.e.,
L n + n M n = 0

and

L m + n M m = 0

where n 6= m and n 6= m .
Multiplying the first equation by m , the second by n , subtracting the resulting
equations and integrating the final expression on [a, b], one obtains
Z

Z
[m L n n L m ] dx + n

Z
m M n dx m

n M m dx = 0
a

Since the system of differential operators and boundary conditions is self-adjoint, and
and since n 6= m , then the integral:
(
Z b
0
n 6= m
(4.24)
m M n dx =
Nn
n=m
a
is a generalized form of an orthogonality integral, with Nn being the normalization
constant.

(ii) Real Eigenfunctions and Eigenvalues


If the system is self-adjoint and positive definite, then the eigenfunctions are real and
the eigenvalues are real and positive. Assuming that a pair of eigenfunctions and
eigenvalues are complex conjugate, i.e.,
n = u(x) + i v(x)

n = n + i n

n = u(x) i v(x)

n = n i n

89

then the orthogonality integral (4.24) results in the following integral:

( )

m M n dx = 0

Since the eigenvalues are complex, i.e., n 6= 0, then


Z b
m M n dx = 0
a

which results in the following real integral:


Z b
(un M un + vn M vn ) dx = 0
a

Recalling the definition of a positive definite system, both of these integrals


are positive, which indicates that the only complex eigenfunction possible is the null
function, i.e., un = vn = 0. One can also show that the eigenvalues n are real and
positive. Starting with the differential equation satisfied by either n and n , i.e.,
Ln + n Mn = 0
and multiplying this equation by n and integrating over [a, b], one obtains an expression for n :
Rb
Rb
L

dx

(un L un + vn L vn ) dx
n
n = n + i n = R ba m
= R ba
M n dx
(un M un + vn M vn ) dx
a m
a
Since the system is positive definite and the integrands are real, then these integrals
are real, which indicates that n = 0 and n is real. Since the system is positive
definite, then the eigenvalues n are also positive. Having established that the eigenvalues of a self-adjoint positive definite system are real and positive, one can obtains
a formula for n . Starting with the equation satisfied by n :
Ln + n Mn = 0
and multiplying the equation by n and integrating the resulting equation on [a, b],
one obtains
Rb
n L n dx
n = R ba
>0
(4.25)

dx
n
n
a

90

(iii) Rayleigh Quotient


The eigenvalues n obtained from (4.25) require the knowledge of the exact form of
the eigenfunction n (x), which of course could have been obtained only if n is already
known. However, one can obtain an approximate upper bound to these eigenvalues
if one can estimate the form of the eigenfunction.
Define the Rayleigh quotient R(u) as
Rb
R(u) = R ba
a

u Lu dx

(4.26)

u Mu dx

where u is a non-vanishing comparison function. It can be shown that for a self-adjoint


and positive definite system:
1 = min R(u)
where u runs through all possible non-vanishing comparison functions. It can also be
shown that if u runs through all possible comparison functions that are orthogonal
to the first r eigenfunctions, i.e.,
Z b
u Mi dx = 0
i = 1, 2, 3, , r
a

then
r+1 = min R(u)
Example:
Obtain approximate values of the first two eigenvalues of the following system:
y 00 + y = 0

0x

y(0) = 0

y() = 0

d
For this system L = dx
2 and M = 1 and hence the system is self-adjoint and also
positive definite. Solving the problem exactly, one can show that it has the following
eigenfunctions and eigenvalues:
)
n (x) = sin(nx)
n = 1, 2, 3,
n = n2

Using the definition of L and M, one can show that Rayleighs quotient becomes:
R b 00
Rb 0 2
u u dx
(u ) dx
a
R(u) = R b
= aR b
u2 dx
u2 dx
a
a
91

where min[R(u)] = 1 = 1.00


One can choose the following comparison functions which satisfies u(o) = u() =
0 and has no other null between 0 and , approximating 1 (x):
(
x/
0 x /2
u1 (x) =
1 x/
/2 x
which is not a proper comparison function, because u0 is discontinuous. The Rayleigh
quotient gives:
R /2
R
(1/)2 dx + (1/)2 dx
12
0
= 2 = 1.23 > 1.00
R1 (u) = + R x/
R

(1/)2 dx + (1 x/)2 dx
0
If one was to use a comparison function that is at least once differentiable, again
approximating 1 (x) such as
u1 (x) = x( x)
R
( 2x)2 dx
10
0
R
= 2 = 1.03 > 1.00
R1 (u) =
2
2

x ( 2x) dx
0
which represents an error of 3 percent.
It can be seen that R1 (u) > 1 = 1, i.e., it is an upper bound to 1 and that the
closer u1 comes to sin x, the close the Rayleigh quotient approaches 1 .
To obtain an approximate value for 2 = 4.00, one can use a comparison function
u2 (x) that has one more null than u2 (x), e.g.,
0 x /4

u2 (x) = 4x/
= 2 4x/

/4 x 3/4

= 4 + 4x/

3/4 x

whose u0 is not continuous. Substituting u2 (x) into R(u), one obtains:


R2 (u) = 4.86 > 4.00
which has a 21 percent error.
Using a comparison function which is at least once differentiable, e.g.,
u2 (x) = x(/2 x)

0 x /2

= (x /)(x //2)

92

/2 x

then the quotient gives:


R2 (u) = 4.053 > 4.00
One should note that the error is down to 1.3 percent for a comparison function which
is at least once differentiable.

93

4.7

Sturm-Liouville System

The Sturm-Liouville(S-L) system is a special case of (4.18) limited to a second order


eigenvalue problem.
Starting with a general, second order operator with an arbitrary parameter:
a0 (x)y 00 + a1 (x)y 0 + a2 (x)y + a3 (x)y = 0

axb

(4.27)

then one rewrite (4.27) in a self-adjoint form by using a multiplier function to the
differential equation int eh form:
(x) =
where

p(x)
a0 (x)

Z
p(x) = exp

a1 (x)
dx
a0 (x)

then the differential equation can be rewritten in the form:


0

[p(x) y 0 ] + q(x) y + r(x)y = 0


where
q(x) =

a1 (x) p(x)
a0 (x)

r(x) =

axb

(4.28)

a3 (x) p(x)
a0 (x)

The two general boundary conditions that can be imposed on y(x) may make the
form:
1 y(a) + 2 y(b) + 3 y 0 (a) + 4 y 0 (b) = 0
1 y(a) + 2 y(b) + 3 y 0 (a) + 4 y 0 (b) = 0
The differential equation (4.28) is self-adjoint, i.e., the operators:


d
d
p
+q
and
M = r(x) are self-adjoint
L=
dx
dx
In order that the system has orthogonal eigenfunctions and positive eigenvalues, the
problem must be self-adjoint and positive definite. The problem is self-adjoint if:
Z
a

b
b
 





0 0
0 0
0
0
u (p v ) + qv v (p u ) + qu dx = P (u, v) = p(x) [uv vu ]
a

= p(b) [u(b)v (b) u (b)v(b)] p(a) [u(a)v (a) u (a)v(a)] = 0 (4.29)

94

Eliminating in turn y(a) and y 0 (a) from the boundary conditions, one obtains:
13 y(a) + 23 y(b) 34 y 0 (b) = 0
13 y 0 (a) + 12 y(b) + 14 y 0 (b) = 0

(4.30)

Eliminating in turn y(b) and y 0 (b) from the boundary conditions, one obtains:
24 y(b) + 14 y(a) 34 y 0 (a) = 0
24 y 0 (b) + 12 y(a) + 23 y 0 (b) = 0

(4.31)

where
ij = i j j i = ij

i, j = 1, 2, 3, 4

If one substitutes for y(a) and y 0 (a) from (4.30) and (4.31) into the self-adjoint condition (4.29), one obtains


24
p(b) p(a)
[u(b) v 0 (b) u0 (b) v(b)] = 0
13
which can be satisfied if
24 p(a) = 13 p(b)

(4.32)

where the identity:


14 23 + 34 12 = 13 24

(i) If 13 = 0, then 24 = 0, and (4.30) and (4.31) becomes:


34 0
y (b) = 0
23
14 0
y(b) +
y (b) = 0
12

34 0
y (a) = 0
14
23 0
y(a)
y (a) = 0
12

y(b)

y(a) +

which indicates that:


34
14
=
12
12
Denoting the ratios 1 =
conditions become:

23
12

34
23
=
23
12

and

> 0, and 2 =

14
12

14
23

1 > 0, then the boundary

y(a) 1 y 0 (a) = 0
y(b) + 2 y 0 (b) = 0

95

(4.33)

In particular,
if 1

and 2 = 0

then

y(a) = 0 and y(b) = 0

if 1

and 2

then

y 0 (a) = 0 and y 0 (b) = 0

if 1 = 0 and 2

then

y(a) = 0 and y 0 (b) = 0

if 1 and 2 = 0

then

y 0 (a) = 0 and y(b) = 0

(4.34)

(ii) If 13 6= 0, then the boundary condition (4.30) and (4.31) can be written as follows:
23
13
12
3 =
13

y(a) = 1 y 0 (b) + 2 y 0 (b)

1 =

y 0 (a) = 3 y 0 (b) + 4 y 0 (b)

34
13
14
and 4 =
13
and 2 =

(4.35)

such that the condition of self-adjointness (4.32) becomes:


(1 4 2 3 ) p(a) = p(b)
In particular, if 2 = 3 = 0 and 1 = 4 = 1, then:
y(a) = y(b)
y 0 (a) = y 0 (b)

(4.36)

p(a) = p(b)
The boundary conditions in (4.36) are known as Periodic Boundary Conditions.
(iii) If p(x) vanishes at an end-point, then there is no need for boundary condition at
that end-point, provided that the product:
lim

xa or xa

p y y 0 0

which can be restricted to y being bounded and y y 0 0 at the specific end point(s).
Thus, the S-L system composed of the differential equation (4.28) and any one of the
possible sets of boundary conditions in (4.33)(4.36), is a self-adjoint system.
The eigenfunctions n of the system are thus orthogonal, satisfying the following
orthogonality integral, (4.24), i.e.,

Z b
0
n 6= m
r(x)n (x)m (x) dx =
Nm
a
n=m
96

In order to insure that the eigenvalues are real and positive, the system must be
positive definite. Thus,
Z b
Z b h
Z b
i
 0 0

2
u Lu dx =
u (pu ) + qu dx =
u p(x) (u0 ) + q(x)u2 dx < 0
a
a
a
Z b
Z b
r u2 dx > 0
u Mu dx =
a

Thus, it is sufficient (but not necessary) that the function p, q, and r satisfy the
following conditions for positive-definiteness:
p(x) > 0
q(x) 0

a<x<b

(4.37)

r(x) > 0
to guarantee real and positive eigenvalues.
It can be shown that the set of orthogonal eigenfunctions of the proper S-L
system with the conditions imposed on p, q, and r constitute a complete orthogonal
set and hence may be used in a Generalized Fourier series.

97

Example: Longitudinal vibration of a free bar


Obtain the eigenfunction and the eigenvalues for the longitudinal vibration of
a free bar, giving explicitly the orthogonality conditions and the normalization constants.
y 00 + y = 0

0xL

y 0 (0) = 0

y 0 (L) = 0

The system is S-L form already,, since it can readily rewritten as:
 
d dy
+ y = 0
dx dx
where
p=1

q=0

and r = 1

The system is a proper S-L system since the differential equation, boundary conditions
are well as the conditions on p, q, and r are those of a proper S-L system;
 
 
y(x) = C1 sin
x + C2 cos
x
y 0 (0) = C1 = 0

 
L = 0
y (L) = C2 sin
0

Thus, the characteristic equation becomes:


sin = 0
having roots n = n,

where

n = 0, 1, 2, .

n2
(n )2
=
n = 0, 1, 2,
L2
L2
The eigenfunction becomes:
 x
n = 0, 1, 2,
n (x) = cos n
L
Note that 0 = 0 is an eigenvalue corresponding to 0 = 1.
n =

The orthogonality condition becomes


Z L
 x
 x
(1) cos n
cos m
dx = 0
n=0
L
L
0
and the normalization becomes

Z L
L
h  x i Z L
 x
 n 
2
2
Nn = N cos n
=
(1)cos n
dx ==
cos
dx = 2
L
L
L
L
0
0
which can be written as N =
2 for n 1.

L
n

n1
n=0

where the Neumann constant is n = 1 for n = 0 and

98

Example: Vibration of a Stretched String with Variable Density


A vibrating stretched string is fixed at x = 0 and x = L, whose density is
varies as:
0 x2
(x) =
L2
The differential equation governing the motion of the string can be written as:
d2 y
0 x2 2
+
y=0
dx2 T0 L2

0xL

with the boundary conditions:


y(0) = 0
Let

0 2

T0

y(L) = 0

= , then the differentia equation becomes


y 00 + x2 y/L2 = 0

The system is in S-L form, with:


p(x) = 1 > 0

q(x) = 0

and r(x) = x2 /L2 > 0

which indicates that it is a proper S-L system.


Referring to (3.45), the solution to the differential equation can be written in
terms of Bessel functions of fractional order:





x2
x2

2 + C2 J1/4
2
y(x) = x C1 J1/4
2L
2L
Since


x2

lim xJ1/4
2 lim x
x0
x0
2L


x2

lim xJ1/4
2 lim x
x0
x0
2L

2 !1/4
x
0
2L2
2 !1/4
!1/4
x

=
2
2L
2L2

then both homogeneous solutions are finite at x = 0. Satisfying the first boundary
conditions yields C2 = 0 and satisfying the second boundary condition yields:
!

y(L) = 0 = C1 J1/4
=0
2
99

which results in the following characteristic equation:

J1/4 () = 0

where

The number of the roots n of the preceding transcendental equation are infinite with
0 = 0 being the first root. Thus, the eigenfunctions and the eigenvalues become



x2
n (x) = xJ1/4 n 2
n = 0, 1, 2,
L
4 2
n = 4n
n = 0, 1, 2,
L

where the number 0 = 0 is not an eigenvalue.


The orthogonality integral is defined as:




Z L 3
Z L 2
x
x
x2
x2
n (x) n (x) dx =
J1/4 m 2 J1/4 n 2 dx = 0
L2
L2
L
L
0
0
and the norm is:
Z
N [n (x)] =
0

x3
L2

2
J1/4



L2
x2
m 2 dx = J3/4 (n )J5/4 (n )
L
4

100

n 6= m

4.8

Sturm-Liouville System for Fourth Order


Equations

Consider a general fourth order linear differential equation of the type that governs
vibration of beams:
a0 (x) y (iv) + a1 (x) y 000 + a2 (x) y 00 + a3 (x) y 0 + a4 (x) y + a5 (x) y = 0,

axb

It can be shown that for this equation to be self-adjoint, the following equalities must
hold:
a1 = 2 a00
a02 a3 = a000
0
It can also be shown that there is no single integrating function that can render this
equation self-adjoint, as was the case of a second order differential operator. Assuming
that these relationships hold and denoting:
 Z x

a1 ()
1
s(x) = exp
d
2
a0 ()
Z x
a3 ()
p(x) =
s() d
a0 ()
then the fourth order equation can be written in self-adjoint form as:
00

Ly + My = (s y 00 ) + (p y 0 ) + (q + r) y = 0

(4.38)

where
a4 (x)
s(x)
a0 (x)
a5 (x)
r(x) =
s(x)
a0 (x)

q(x) =

For the fourth order S-L system to have orthogonal and real eigenfunctions and
positive eigenvalues, the system must be self-adjoint and positive definite. In the
notation of (4.18), the operators L and M are:




d2
d2
d
d
L =
s(x) 2 +
p(x)
+ q(x)
dx2
dx
dx
dx
M = r(x)

101

The system is self-adjoint, so that P (u, v) given by (4.22), is given by


b
b
P (u, v) a = u(s v 00 )0 v(s u00 )0 s(u0 v 00 u00 v 0 ) + P (u v 0 v u0 ) a = 0

(4.39)

Boundary conditions on y, and consequently on the comparison functions u and v,


can be prescribed such that (4.39) is satisfied identically. The five pairs of boundary
conditions are listed below:

(i) y = 0

y0 = 0

(ii) y = 0

s y 00 = 0

(iii) y 0 = 0

(s y 00 ) = 0

(vi)

(s y 00 ) y = 0

(4.40)

y0 = 0

(v) s y 00 y 0 = 0

y=0

where + sign for x = b and for x = a.


If p(a) or p(b) vanishes (singular boundary conditions), then at the end point
where p(x) vanishes, the boundedness condition is invoked, i.e.,
lim p y y 0 0

xa or b

(which can be restricted to y being finite and p y 0 0), as well as the following
pairs of boundary conditions in addition to those given in (4.40), can be specified at
the end where p = 0:
(i)
(s y 00 )0 = 0
s y 00 = 0
(ii) (s y 00 )0 y = 0
s y 00 = 0
(iii) s y 00 y = 0
(s y 00 )0 = 0
(iv)
y=0
y=0
(v) (s y 00 )0 y = 0 s y 00 y = 0

(4.41)

where + sign for x = b and for x = a.


If s(x) vanishes at one end(singular boundary conditions), then together with
the requirement that:
lim s y 0 y 00 0
xa or b

the following boundary conditions can be prescribed at the end where s(x) vanishes:
(i)
y=0
s y0 = 0
(ii)
(s y 00 )0 = 0
s y0 = 0
(iii) (s y 00 )0 y = 0
y0 = 0
(iv)
y=0
y=0
(v)
(s y 00 )0 = 0
y0 y = 0
102

(4.42)

where + sign for x = b and for x = a, respectively.


If both p(x) and s(x) vanish at one end(singular boundary conditions), then
together with the requirement that:
lim p y y 0 0

xa or b

lim s y 0 y 00 0

xa or b

and

lim s y y 000 0

xa or b

one can prescribe the following condition at the end where p(x) and s(x) vanish having the form:

(i) y = 0
(ii) (s y 00 )0 = 0

(4.43)

(iii) (s y 00 )0 y = 0
where + sign for x = b and for x = a, respectively.
If p(x), s(x) and s0 (x) vanish at one end point, then there are no boundary
conditions at those ends provided that:
lim s y y 000 0

lim s0 y y 00 0

xa or b

xa or b

and
lim s y 0 y 00 0

lim p y y 000 0

xa or b

xa or b

If p(x) = 0 in a x b, then the nine boundary conditions specified in (4.40)


and (4.41) satisfy (4.39), as was shown for beam vibrations.
More complicated boundary conditions of the type:
i1 y 000 (a) + i2 y 00 (a) + i3 y 0 (a) + i4 y(a)
+ i1 y 000 (b) + i2 y 00 (b) + i3 y 0 (b) + i4 y(b) = 0

i = 1, 2, 3, 4

can be postulated to develop the conditions on ij and ij under which such boundary
conditions satisfy (4.39).
To guarantee positive eigenvalues, the system must be positive definite. Then,
the following inequalities must hold:
Z b
Z bh
i


2
2
00 00
0 0
u (u u ) + (p u ) + q u dx =
q u2 p (u0 ) + s (u00 ) dx < 0
a

and
Z

Z
u r u dx =

103

r u2 dx > 0

where the boundary conditions specified in (4.40)(4.43) were used. Thus, sufficient
(but not necessary) conditions on the functions can be imposed to satisfy positive
definiteness:
p 0
r > 0
q 0

a<x<b

s < 0

4.9

Solution of Non-Homogeneous Eigenvalue Problems

Consider the following non-homogeneous system:


Ly + My = F (x)

axb

(4.44)

Ui (y) = i

i = 1, 2, , 2n

(4.45)

where L and M are self-adjoint operators and Ui were given in (4.18) and (4.19) and
is a given constant.
Due to the linearity of the system in (4.45), one can split the solution into
two parts. The first solution satisfies the homogeneous differential equation with
non-homogeneous boundary conditions and the second system satisfies the nonhomogeneous equation with homogeneous boundary condition. The sum of the two
solutions satisfy the original system of (4.45).
Let y = yI (x) + yII (x) such that
LyI + MyI = 0

LyII + MyII = F (x)

Ui (yI ) = i

Ui (yII ) = 0

i = 1, 2, , 2n

(4.46)

(i) The solution yI (x) in (4.46) can be obtained by solving the homogeneous differential equation on yI (x) and substituting the (2n) independent solutions into the
non-homogenous boundary conditions for yI (x).
yI (x) = c1 y1 (x) + c2 y2 (x) + + c2n y2n (x)

(4.47)

where the arbitrary constants {ci , i = 1, 2, , 2n} are determined using the boundary conditions in (4.46). It should be noted that if i 0, then yI (x) 0.
104

(ii) The solution yII (x) in (4.46) can be developed by utilizing the eigenfunctions of
the following homogeneous system:
Lm + m Mm = 0

(4.48)

Ui (m ) = 0

(4.49)

i = 1, 2, , 2n

The eigenfunctions m (x) of the system in (4.48) must be obtained first, where each
eigenfunction satisfies the homogeneous boundary conditions in (4.49).
The set of eigenfunction {m (x)} satisfy the orthogonality integral (4.24). The
solution yII (x) can be expanded in a generalized Fourier series in the eigenfunction of
(4.48) as follows:

X
an n (x)
(4.50)
yII =
n=1

Substituting the series in (4.50) into the differential equation on yII (x), one obtains

an Ln +

n=1

an Mn = F (x)

(4.51)

n=1

Substituting for Ln from (4.48) into (4.51), one obtains:

[( n ) an Mn ] = F (x)

(4.52)

n=1

Multiplying both sides of (4.52) by m (x), integrating over [a, b] and invoking the
orthogonality relationship in (4.24), one obtains:
an =

bn
( n ) Nn

where Nn is the norm of the eigenfunctions


(
Z
b

m M n dx =
a

and
Z

0
Nn

n 6= m
n=m

bn =

F (x)n (x) dx

(4.53)

bn
n (x)
( n ) Nn

(4.54)

Thus, the solution to yII (x) becomes


yII (x) =

b
X
a

105

The solution due to the source term F (x) can be seen to become unbounded whenever
becomes equal to any of the eigenvalues n .
Hence, the solution to the differential equation in (4.45) is obtained by summing
yI (x) and yII (x) as
y(x) = yI (x) + yII (x)
where yI (x) is in (4.47) and yII (x) is (4.54).

Example: Forced Vibration of a Simply-Supported Beam

P0
2a

-a

L/2

Obtain the steady state deflection of a simply supported beam being vibrated
by a distributed load as follows:
f (x) = f (x) sin( t)
where

P /2,
0
f (x) =
0

L/2 a < x < L/2 + a


everywhere else

The beam has a length L and has a constant cross-section. It is simply supported at
both ends such that:
y (0, t) = 0

y 00 (0, t) = 0

y (L, t) = 0

y 00 (L, t) = 0

Letting y (x, t) = y(x) sin(t), then:


y (iv) + 4 y =

f (x)
EI

where
106

4 =

A 2
EI

where the boundary conditions are


y(0) = 0

y 00 (0) = 0

y(L) = 0

y 00 (L) = 0

(i) The first system for yI (x) is obtained as


(iv)

yI

+ 4 yI = 0

4 =

where

A 2
EI

with
yI (0) = 0

yI00 (0) = 0

yI (L) = 0

yI00 (L) = 0

Since the differential equation is homogeneous and the boundary conditions are homogeneous, the solution yI (x) is a trivial solution, that is, yI (x) 0.
(ii) The second system for yII (x) is obtained as
(iv)

yII + 4 yII =

f (x)
EI

4 =

where

A 2
EI

with
yII (0) = 0

yII00 (0) = 0

yII (L) = 0

yII00 (L) = 0

One must find the eigenfunctions of the system first:


u(iv) + u = 0

where = 4

L=

d4
dx4

M =1

with
uII (0) = 0

u00II (0) = 0

uII (L) = 0

u00II (L) = 0

The solution of the fourth order differential equation with constant coefficients is:
u = C1 sin x + C2 cos x + C3 sinh x + C4 cosh x
Satisfying the boundary conditions:
u(0) = 0 = C2 + C4 = 0
u00 (0) = 0 = C2 + C4 = 0
107

which means that C2 = C4 = 0.


u(L) = 0 = C1 sin L + C3 sinh L = 0
u00 (L) = 0 = C1 sin L + C3 sinh L = 0
which results in C3 = 0. The characteristic equation becomes:
sin = 0

where = L

which has roots n = n , n = 0, 1, 2, . The zero root results in a zero solution, so


that 0 = 0 is not an eigenvalue.
The corresponding eigenfunctions become:
 x
 n 
n (x) = sin n
= sin
x
L
L
n4 4
4
n = n4 = n4 = 4
L
L
The orthogonality condition is given by

Z L
0
 m 
 n 
x sin
x dx =
sin
L/2
L
L
0

n = 1, 2,
n = 1, 2,

n 6= m
n=m

Since the boundary conditions are homogeneous, then yI = 0 and y = yII . Expanding
the function y(x) into an infinite series of the eigenfunctions, then the constant bn is
given by:



Z L/2+a 
 n 
a
sin n
sin n
P0
2
L
bn =

sin
x dx = P0
2a
L
EI (na/L)
L/2a
Thus, the solution becomes:


sin
2P0 X sin n
2

y(x) =
4
EI L n=1 4 nL4 4

n
a
L

na
L

sin

 n 
x
L

If a 0, the distributed forcing function becomes a concentrated force, P0 , then the


limit of the solution approaches:


 n 
2P0 X sin n
2

y(x)a0
sin
x
EI L n=1 4 nL4 4 4
L
For concentrated point sources and forces, one can represent them by Dirac delta
functions. Thus, one can represent f (x) by:
f (x) = P0 (x L/2)
The constant bn can now be found using the sifting property of Dirac delta function:
Z L
 n 
 n 
bn = P0
(x L/2) sin
x dx = P0 sin
L
2
0

108

Вам также может понравиться