The Notebook Series

The series solution of second order,

ordinary differential equations and
special functions



by



R.S. Johnson

Professor of Applied Mathematics

School of Mathematics & Statistics
University of Newcastle upon Tyne













R.S.Johnson 2006
2
CONTENTS
List of equations 3
Preface . 4
1. Power-series solution of ODEs 5
1.1 Series solution; essential ideas 5
1.2 ODEs with regular singular points 8
Exercises 1 .. 10
2. The method of Frobenius 11
2.1 The basic method .. 11
2.2 Two special cases .. 13
Exercises 2 .. 24
3. The Bessel equation and Bessel functions 25
3.1 First solution 25
3.2 The second solution . 27
3.3 The modified Bessel equation 29
Exercises 3 31
4. The Legendre polynomials 33
Exercises 4 .. 34
5. The Hermite polynomials . 36
Exercises 5 . 37
6. Generating functions. 38
6.1 Legendre polynomial 38
6.2 Hermite polynomials 40
6.3 Bessel functions .. 41
Exercises 6 .. 43
Answers . 44
Index 46






3
List of Equations

This is a list of the types of equation, and specific examples, whose solutions are
discussed. (Throughout, we write y y x = ( ) and a prime denotes the derivative; the
power series are about x = 0, unless stated otherwise.)

+ + = + y y y x 3 2 2 3 ..... p.6
a x y b x y c x y ( ) ( ) ( ) + + = 0 regular singular points (general) ... p.8
Classify singular points:
+ + = y y y 3 2 0 ; x y xy y
2
3 2 0 + + = ; x x y xy y ( ) 1 0 + + = ;
x y x y x y
3 2
1 0 + + = ( ) . all on p.9
x y xy y
2
3 2 0 + + = classify point at infinity .... p.10
4 2 0 xy y y + = .. p.12
xy y xy + = 0 ... p.15
xy y + = 0 ..... p.17
( ) 1 2 0 + + = x y xy y about x = 1 ...p.19
x x y xy x y
2 1
4
1 3 0 ( ) ( ) + = about x = 0 and x = 1 .. p.20
x y xy x y
2 2 2
0 + + =
e j
(Bessels equations) ...... p.25
xy y a xy + + =
2
0 ..... p.29
x y xy x y
2 2 2
0 + + =
e j
(modified Bessel equation) ... p.29
1 2 0
2
+ = x y xy y
e j
(Legendres equation) .. p.33
+ = y xy y 2 0 (Hermites equation) .... p.36





















4
Preface

This text is intended to provide an introduction to the methods for solving
second order, ordinary differential equations (ODEs) by the method of Frobenius. The
topics covered include all those that are discussed in the degree programmes offered
in the School of Mathematics & Statistics at Newcastle University. The material has
been written to provide a general introduction to the relevant ideas, rather than as a
text linked to a specific module. Indeed, the intention is to present the material in a
way that enhances the understanding of the topic, and so can be used as an adjunct to
a number of different modules or simply to help the reader gain a broader
experience of mathematics. The aim is to go beyond the methods and techniques that
are presented in our modules, but all the standard ideas are included (and can be
accessed through the comprehensive index).

It is assumed that the reader has a basic knowledge of, and practical
experience in, various aspects of the differential and the integral calculus. In
particular, familiarity with the basic calculus and elementary differential-equation
work of Stage 1 is assumed. This brief notebook does not attempt to include any
applications of the differential equations; this is properly left to a specific module that
might be offered in a conventional applied mathematics or engineering mathematics
or physics programme. However, the techniques are applied to some specific
equations whose solutions include important special functions that are met in most
branches of mathematical methods; thus we will discuss: Bessel functions, and the
polynomials of Legendre and Hermite.

The approach adopted here is to present some general ideas, which might
involve a notation, or a definition, or a theorem, or a classification, but most
particularly methods of solution, explained with a number of carefully worked
examples (there are 11 in total). A small number of exercises, with answers, are also
offered, although it must be emphasised that this notebook is not designed to be a
comprehensive text in the conventional sense.


Robin Johnson, September 2006















5
1. Power-series solution of ODEs

In this chapter we will describe the fundamental ideas and method that underpin
this approach to the solution of (second order, linear) ordinary differential equations.
This will be presented with the help of a simple example, which will provide much of
the motivation for the more general methods that follow later. However, as we explain
in 1.2, this first analysis is very restrictive. We extend its applicability by first
classifying the equations for which we can employ this technique, and then we
carefully formulate (Chapter 2) the general method (due to G. Frobenius). All the
possible cases will be described, with examples, and then we apply the procedure to a
number of important equations.

1.1 Series solution: essential ideas

The methods for finding exact solutions of ordinary differential equations (ODEs)
are familiar; see, for example, the volume The integration of ordinary differential
equations in The Notebook Series. So, for example, the equation

+ + = + y y y x 3 2 3 2

(where the prime denotes the derivative with respect to x) has the complete general
solution (complementary function + particular integral)

y x A B x
x x
( ) = + +

e e
2
,

where A and B are the arbitrary constants. A solution expressed like this is usually
referred to as being in closed form; on the other hand, if we wrote the solution (of
some problem) in the form

y x A
x
n
x
n
n
( ) =
F
H
G
I
K
J
+
=

2
1
2


then this is not in closed form. (It would become closed form if we were able to sum
the series in terms of elementary functions.) A solution is, of course, best written in
closed form, but we may not always be able to do this; it is, nevertheless, sufficient
for most purposes to represent the solution as a power series (provided that this series
is convergent for some x, so that the solution exists somewhere). We should note that
the solution of
+ + = + y y y x 3 2 3 2

could be written

y x A
x
n
B
x
n
x
n
n
n
n
( )
( )
!
( )
!
=

F
H
G
I
K
J
+

F
H
G
I
K
J
+
=

0 0
2


(with the usual identification: 0! 1 = ). In other words, we could always seek a power-
series solution, and this will be particularly significant if we cannot solve the equation
6
any other way. Indeed, it is evident that this approach provides a more general
technique for tackling the problem of solving differential equations, even if the
downside is the construction of a more complicated-looking form of the solution.
Thus the procedure is to set y x a x
n
n
n
( ) =
=

0
, and then aim to determine the
coefficients, a
n
, of the series, in order to ensure that this series is the solution of the
given equation.

Example 1
Seek a solution of + + = + y y y x 3 2 2 3 in the form y x a x
n
n
n
( ) =
=

0
.

Given the power series, we find

=

=

y na x
n
n
n
1
0
and =

=

y n n a x
n
n
n
( ) 1
2
0


and so the equation becomes

n n a x na x a x x
n
n
n
n
n
n
n n n
( ) + + = +

=

1 3 2 3 2
1 1
0 0 0
.

We have assumed that y, y and y , all expressed via the given series, exist for some
x i.e. all three series are convergent for some xs common to all three series. With this
in mind, we require the equation, expressed in terms of the series, to be valid for all x
in some domain so we do not generate an equation for x! For this to be the case, x
must vanish identically from the equation. Now our equation, written out in more
detail, becomes

2 6 12 20 3 2 3 4
2 3 4
2
5
3
1 2 3
2
4
3
a a x a x a x a a x a x a x + + + + + + + + + ... ...
e j

+ + + + + = + 2 3 2
0 1 2
2
3
3
a a x a x a x x ...
e j


and so, to be an identity in x, we require

2 3 2 3
2 1 0
a a a + + = ; 6 6 2 2
3 2 1
a a a + + = ;
12 9 2 0
4 3 2
a a a + + = ; 20 12 2 0
5 4 3
a a a + + = , and so on.

We choose to solve these equations in the form

a a a
2
3
2
3
2
1 0
= ; a a a a a
3
1
3
1 2
7
6
1 0
1 1 = = + ( ) ( ) ;
a a a a a
4
1
6
2
3
4
3
5
8
1
7
12
0
1 = = ( ) ; a a a a a
5
1
10
3
3
5
4
31
120
1
1
4
0
1 = = + ( ) , etc.

7
Further, see that, in general, for every term x
n
in the equation, for n 2 , we may
write
( )( ) ( ) n n a n a a
n n n
+ + + + + =
+ +
1 2 3 1 2 0
2 1


i.e. a
a
n n
a
n
n
n n
+
+
=
+ +

+
2
1
2
1 2
3
2 ( )( )
;

this is called a recurrence relation.

Because we have the combination (1
1
a ) appearing here, it is convenient to write
a b
1 1
1 = + , then all the coefficients a
2
, a
3
, etc., depend on only two constants: a
0

and b
1
. These are undetermined in this system, so they are arbitrary: the two arbitrary
constants expected in the general solution of a second order ODE. Our solution
therefore takes the form

y x a b x a b x a b x a b x ( ) ( ) = + + + + + +
0 1 0
3
2
1
2
0
7
6
1
3 7
12
0
5
8
1
4
1
e j e j e j

+ + +
1
4
0
31
120
1
5
a b x
e j
....

= + + + + x a x x x x
0
2 3 7
12
4 1
4
5
1 ...
e j

+ + + + b x x x x x
1
3
2
2 7
6
3 5
8
4 31
120
5
...
e j
.

This is more conveniently written by relabelling the arbitrary constants as

a A B
0
= + ; b A B
1
2 = + ( )

although this is certainly not a necessary manoeuvre. (We choose to do this here to
show directly the connection with the general solution quoted earlier.) This gives

y x A x x x x x ( ) ...
! ! ! !
= + + + 1
1
2
2 1
3
3 1
4
4 1
5
5
e j

+ + + + + B x x x x x x 1 2 2 2 2 2
1
2
2 1
3
3 1
4
4 1
5
5
! ! ! !
( ) ( ) ( ) ( ) ...
e j
,

which recovers our original (closed form) solution, because the infinite series (on
using the formula that defines all the coefficients) can be summed and are convergent
for all finite x. This approach, we observe, has automatically generated both the
complementary function and the particular integral.


In summary, direct substitution of a suitable series, followed by the identical
elimination of x, enables the complete general solution to be found. This approach, we
note perhaps with some relief does not invoke any integration processes; any
technical difficulties are now associated with the issue of convergence of the resulting
series.

8
1.2 ODEs with regular singular points

The essential idea described in 1.1 needs to be developed further because it
cannot accommodate, as it stands, functions such as y x x x ( ) ( ) = + 1 . This function,
represented as a power series, becomes
y x x x x ( ) ... = + + 1
1
2
1
8
2
e j
, x <1,

obtained by using the familiar binomial theorem; this contains non-integral powers of
x. Thus a solution assumed to be of the form used in 1.1:

y x a x
n
n
n
( ) =
=

0
,

can never find a solution that contains a term such as x . The method a simple
extension of our basic approach will enable a wide class of important equations to
be solved. However, we must be clear about this class, and this then leads to an
important classification of linear, second order, ordinary differential equations.

Let us suppose that the homogeneous ODE

a x y b x y c x y ( ) ( ) ( ) + + = 0,

where the coefficients a, b and c are given functions, possesses a solution, near some
point x x =
0
, of the form

y x A x x ( )
0
b g

(as x x
0
)

for some constant (where A is an arbitrary constant). Then we see that



y A x x

0
1
b g ;

y A x x

( ) 1
0
2
b g ,

and so y , y x x ( )
0
and y x x ( )
0
2
are each proportional to ( ) x x

0
2
. Thus
a linear combination of these will be zero, and this must recover (approximately) the
original ODE (if such a solution exists) in the form

+ + = y
b x
a x
y
c x
a x
y
( )
( )
( )
( )
0

i.e. b x a x x x ( ) ( ) ( )

0
1
and c x a x x x ( ) ( ) ( )

0
2
(as x x
0
).

This observation is the basis for the classification of the equations that we are able to
solve by this method. Note that we are not concerned, at this stage, with the complete
general solution which contains a particular integral.

Consider the second-order, homogeneous, linear ODE, written in the form
9

+ + = y p x y q x y ( ) ( ) 0 ;

if p x ( ) or q x ( ) (or both) are undefined in this context, this means infinite at
x x =
0
, then we have a singular point at x x =
0
. Then, in the light of the observations
just made, we form

P x x x p x ( ) ( ) ( ) =
0
and Q x x x q x ( ) ( ) ( ) =
0
2
;

if both P x ( ) and Q x ( ) remain finite as x x
0
, then we have a regular singular
point at x x =
0
.

If either P x ( ) or Q x ( ) (or both) are undefined (infinite) at x x =
0
, then we have an
essential singular point at x x =
0
. On the other hand, if both p x ( ) and q x ( ) are
defined at x x =
0
, then we have an ordinary point at x x =
0
.

The methods that we shall develop here are applicable to equations with regular
singular (or ordinary) points; we cannot, in general, use this same approach to find a
solution in the neighbourhood of an essential singular point.

Example 2

Identify and classify the singular points of these equations.

(a) + + = y y y 3 2 0 ; (b) x y xy y
2
3 2 0 + + = ;
(c) x x y xy y ( ) 1 0 + + = ; (d) x y x y x y
3 2
1 0 + + = ( ) .

In each, first we identify the functions p x ( ) and q x ( ) , and then proceed.

(a) p x ( ) = 3, q x ( ) = 2 : defined for all x, so all points are ordinary points.

(b) p x x ( ) = 3 , q x x ( ) = 2
2
: there is a singular point at x = 0; we form xp x ( ) = 3
and x q x
2
2 ( ) = , both of which are defined at x = 0, so we have a regular singular
point at x = 0.

(c) p x x ( ) ( ) = 1 1 , q x x x ( ) ( ) = 1 1 : there are singular points at x = 0 and x = 1; for
x = 0 we form xp x x x ( ) ( ) = 1 and x q x x x
2
1 ( ) ( ) = , both of which are defined at
x = 0, so x = 0 is a regular singular point; for x = 1 we form ( ) ( ) x p x = 1 1 and
( ) ( ) ( ) x q x x x = 1 1
2
, both of which are defined at x = 1, so x = 1 is another
regular singular point.

(d) p x x ( ) = 1 , q x x x ( ) ( ) = 1
3
: there is a singular point at x = 0; we form
xp x ( ) = 1 and x q x x x
2
1 ( ) ( ) = , and the second of these is not defined at x = 0, so
we have an essential singular point at x = 0.
10
Comment: This classification can also be applied where x . To accomplish this,
we transform the equation so that y x Y x ( ) ( ) =
1
which ensures that x now
corresponds to x u

=
1
0; u = 0 is called the point at infinity. The point at infinity
will, according to our classification, be an ordinary point or a regular singular point or
an essential singular point.

Example 3

Classify the point at infinity for the equation x y xy y
2
3 2 0 + + = (cf. Ex. 2(b)).

We write y x Y x ( ) ( ) =
1
then
=

y x
x
Y x ( ) ( )
1
2
1
and = +

y x
x
Y x
x
Y x ( ) ( ) ( )
2 1
3
1
4
1
,

so we have y Y u = ( ) , = y u Y u
2
( ) , = + y u Y u u Y u 2
3 4
( ) ( ) (where u x =
1
).
Thus the equation becomes

1
2
3
2 0
2
3 4 2
u
u Y u Y
u
u Y Y + + + =
e j e j


i.e. u Y uY Y
2
2 0 + = .

We see that p u u ( ) = 1 , q u u ( ) = 2
2
: the point at infinity ( u = 0 ) is singular point.
Now consider up u ( ) = 1 and u q u
2
2 ( ) = , both of which are defined at u = 0 , so the
point at infinity is a regular singular point.


Exercises 1

1.Identify and classify the singular points (excluding the point at infinity) of these
equations.
(a) x x y x x y x y
2
1 1 1 0 ( ) ( ) ( ) + + + = ; (b) 1
1
0
2 2
+
+
F
H
G
I
K
J + = x y
x
x
y x y
e j
;
(c) + = y x y
2
0, and in this case, also classify the point at infinity.

2. Show that the equation (the hypergeometric equation)

x x y c a b x y aby ( ) ( ) 1 1 0 + + + = l q ,

has three regular singular points: at x = 0, x = 1 and at infinity (where a, b and c
are given constants, although c will not be zero, nor will c be an integer).

****************
**********
11
2. The method of Frobenius

We consider the second order, linear ordinary differential equation

a x y b x y c x y r x ( ) ( ) ( ) ( ) + + =

for which we know that the solution can be expressed as

y x y x y x
CF PI
( ) ( ) ( ) = + ,

where the particular integral, y y x
PI
= ( ) , is any function that will generate r x ( ) . It is
known that, if we can find one or both solutions that contribute to the complementary
function, y x
CF
( ) , then the complete general solution can be found (by the method of
variation of parameters, for example). Thus it is sufficient here to work only with the
homogeneous equation i.e. r x ( ) 0. (For the relevant background ideas and
techniques, see volume The integration of ordinary differential equations in The
Notebook Series, or any other suitable text.) We shall consider a power-series
solution, in powers of (x x
0
) (and quite often we will have x
0
0 = ) and if x x =
0

happens to be an ordinary point of the equation, then the simple method described in
the previous chapter is applicable. We allow the equation to have, at most, regular
singular points, and then the most useful results are obtained by expanding about a
regular singular point i.e. x x =
0
may be a regular singular point. The method cannot
be applied, in general, if x x =
0
is an essential singular point. If there are, for
example, two regular singular points, then it is often necessary to construct two power
series: one about each of the regular singular points; we will comment on this in more
detail later.

2.1 The basic method

Given the equation
a x y b x y c x y ( ) ( ) ( ) + + = 0,

which has a regular singular point at x x =
0
, we seek a solution

y x a x x
n
n
n
( ) =
+
=

0
0
b g

, a
0
0 ,

where is a parameter (called the index) which is to be determined; this is usually
called a power series about x x =
0
. This type of series solution is at the heart of the
method of Frobenius. [Georg F. Frobenius (1849-1917), German mathematician;
developed the concept of the abstract group; published his method for finding power-
series solution of second order ODEs early in his career (in 1874).] Apart from the
appearance of , the method follows precisely that described in Chapter 1. Note that
gives the power of the first term ( n = 0 ) in the series, whatever that may be. Thus,
provided that a power series of this form exists, there must be a first term and this has
been designated a x x
0 0
b g

, and so we require a
0
0 . As we shall see, our previous
12
method can be simplified and made more systematic, and this is the approach we will
incorporate here.

Example 4

Use the method of Frobenius to find the general solution of the equation

4 2 0 xy y y + = ,

as a power series about x = 0.


We can note that the equation possesses a regular singular point at x = 0, so the
method is applicable. We set y x a x
n
n
n
( ) =
+
=

0
( a
0
0 ), then we obtain

= +
+
=

y x a n x
n
n
n
( ) ( )
1
0
; = + +
+
=

y x a n n x
n
n
n
( ) ( )( )

1
2
0
,

and so the equation becomes

4 1 2 0
2
0
1
0 0
x a n n x a n x a x
n
n
n
n
n
n
n
n
n
( )( ) ( )

+ + + + =
+
=

+
=

+
=

.

It is convenient to introduce new dummy counters into each summation, so that each
term now takes the form x
m +
this is the essential manoeuvre in simplifying and
organising the calculation. Thus in the first and second series we write n m = 1 , and
in the third we simply set n m = ; we then obtain

4 1 2 1 0
1
1
1
1 0
a m m x a m x a x
m
m
m
m
m
m
m
m
m
+
+
=

+
+
=

+
=

+ + + + + + =



b gb g b g .

This equation can be rearranged to give

4 1 2
0
1
0
1
a x a x

( ) +


+ + + + + + + =
+ +
+
=

4 1 2 1 0
1 1
0
a m m a m a x
m m m
m
m


b gb g b g m r


which is an identity, for all x, if

2 2 0
0
2
a ( ) = ;
2 1 2 2 1 0
1
+ + + + =
+
m m a a
m m
b gb g , m= 0 1 2 , , , ... .

But, by definition, a
0
0 so we require
13

2 0
2
= ,

and the equation for is called the indicial equation (being the equation for the index
); here we see that = 0 or 1 2.

We also have
a
a
m m
m
m
+
=
+ + + +
1
2 1 2 2 1 b gb g
( m= 0 1 2 , , , ... )

which is called the recurrence relation. (Note that we are not dividing by zero here,
for the given s and ms but this is something that we must be wary of later.)

Now for each value of , we can find the coefficients a
m
, ultimately in terms of a
0
.
However, because the equation is linear, we may perform the two calculations (one
each for = 0 and = 1 2 ) and then construct a general linear combination of the
two in order to generate the complementary function. In particular, for = 0, we
obtain

a
a
m m
m
m
+
=
+ +
1
2 1 2 1 ( )( )
: a a
1
1
2
0
= ; a a
a
2
1
2.2.3
1
0
2 2
= =
( . )!
, etc.,

and for = 1 2 :

a
a
m m
m
m
+
=
+ +
1
2 2 3 1 ( )( )
: a a
1
1
2.3
0
= ; a a
a
2
1
2.2.5
1
0
2 2 1
= =
+ ( . )!
, etc.

The general solution can then be written as

y x A
x
n
Bx
x
n
n
n
n
n
( )
( )! ( )!
= +
+
=

2 2 1
0
1 2
0
.

Comment: These two series converge for all finite x, as can be seen by applying the
ratio test.

In this example, we found that the indicial equation was quadratic it always will
be and that its roots were not equal (nor did they differ by an integer, which will be
an important observation). The upshot is that we have two special cases that must be
addressed.

2.2 The two special cases

First, we write the general homogeneous equation in the form

+ + = y p x y q x y ( ) ( ) 0

14
and seek a solution
y x a x
n
n
n
( ) =
+
=

0
;

this will, for general , take the form

y x a x x x ( ) ( ) ( ) .... = + + +
0 1 2
2
1

{ }


where the
i
( ) are obtained from the recurrence relation. Let us write

Y x x x x ( , ) ( ) ( ) ....

= + + + 1
1 2
2
{ }
,

then the equation, with y Y = , gives (after a little calculation, which is left as an
exercise)
+ + =

Y p x Y q x Y x ( ) ( )


2
1 2
b gb g ,

where =
1 2
, are the roots of the indicial equation. This result immediately
confirms what happened in Example 4: y Y x = ( , ) is a solution for =
1
and for
=
2
. However, if
1 2
= (repeated root), we clearly obtain only one solution by
this method (and we know that there must be two linearly independent solutions of a
second order ODE). To see how to continue in this case, we choose to differentiate
with respect to :

+ + =

Y p x Y q x Y x

( ) ( )
2
1
2
b g
{ }


where we have set
2 1
= . But the right-hand side of this equation is still zero for
=
1
, so a second solution is

Y x ( , ) evaluated on =
1
. This solution is

+ + +
=

x x x 1
1 2
2
1
( ) ( ) ....
{ }
on

=

+ + +
L
N
M
O
Q
P
=

e
ln
....
x
x x 1
1 2
2
1
{ }

= + + + +

+ + x x x x x x x


1 1
1
1 2
2
1 1 2 1
2
ln .... ( ) ( ) ...
{ } { }


this second solution contains a logarithmic term, which is therefore certainly
linearly independent of our first solution (Y). As we have demonstrated, this second
solution can be obtained directly from the first solution or, as is quite often done, it is
computed by seeking an appropriate solution of the differential equation.



15
Example 5

Use the method of Frobenius to find the general solution of the equation
xy y xy + = 0

as a suitable power series about x = 0.


We note that the equation has a regular singular point at x = 0, so we may indeed
write y x a x
n
n
n
( ) =
+
=

0
( a
0
0 ), to give

a n n x a n x a x
n
n
n
n
n
n
n
n
n
( )( ) ( )

+ + + + =
+
=

+
=

+ +
=

1 0
1
0
1
0
1
0
.

In the first two summations, we choose to write m n = 1, and m n = +1 in the third;
this gives
a m x a x
m
m
m
m
m
m
+
+
=

+
=

+ + =
1
2
1
1
1
1 0

b g .

This can now be expressed as

a x a x a m a x
m m
m
m
0
2 1
1
2
1
2
1
1
1 1 0

+
+
=

+ + + + + =

( ) ( )
{ }


and this is an identity for all x if

a
0
2
0 = ; a
1
2
1 0 ( ) + = ; a m a
m m +
+ + =
1
2
1
1 0 ( ) ( m= 1 2 , , ... ).

But a
0
0 , so = 0 (repeated) and then a
1
0 = ; this leaves

a
a
m
m
m
+

=
+
1
1
2
1 ( )
( m= 1 2 , , ... ),

with implies that 0
1 3 5
= = = = a a a .... . Otherwise we obtain

a
a
2
0
2
2
= ; a
a a a
4
2
2
0
2
0
4 2
4 2 4 2 2
= = =
( . ) ( !)
, etc.,

which provides one solution:

y x a
x
m
n
n
1 0
2
2
0
2
( )
( )
( !)
=
=

,

16
where a
0
is the arbitrary constant.

To proceed, let us write y x a Y x
1 0
( ) ( ) = (so, of course, xY Y xY + = 0), then a
second solution must necessarily take the form

y x Y x x b x
n
n
n
2
0
( ) ( ) ln = +
=

;

the equation therefore gives

x Y x Y
x
Y
x
b n n x
n
n
n
+ +
F
H
G
I
K
J
+
L
N
M
M
O
Q
P
P

ln . . ( ) 2
1 1
1
2
2
0

+ + + +
L
N
M
M
O
Q
P
P
=

Y x Y
x
b nx x Y x b x
n
n
n
n
n
n
ln . ln
1
0
1
0 0
.

Thus, when we use the fact that Y x ( ) is a solution of the original equation, we obtain

b n x b x
nx
n
n
n
n
n
n
n
n
n
n
2 1
1
1
0
2 1
2 2
0
2
2
2
0

+
=

=

+ =
( !)


so 0
1 3 5
= = = = b b b .... and

( )
( )
( )!
2 2
1
2 1
2
2 2 2
2 2
m b b
m
m
m m
m
+ =
+
+
+
, m= 0 1 2 , , , ... ,

(where we have written 2m for n the even values). We may select b
0
0 = (because
this will simply generate y
1
again), then

b
2
1
4
= ; b
4
3
816
=
.
, etc.

The complete, general solution can therefore be written as

y x A
x
n
B
x
n
x x x
n
n
n
n
( )
( )
( !)
( )
( !)
ln
.
... = +
F
H
G
I
K
J
+
R
S
|
T
|
U
V
|
W
|
=

2 2 1
4
3
816
2
2
0
2
2
0
2 4
,

where the arbitrary constants are now A and B.


The second special case arises when the roots for differ by an integer; let us
write
1 2
= + n, where n is a positive integer i.e.
1 2
> . A solution for general
will take the same form as before, namely

17
y x a x x x ( ) ( ) ( ) .... = + + +
0 1 2
2
1

{ }
,

but this time each
i
( ) , starting at
n
( ) (with n as just defined), will have a factor
( )
2
in the denominator. (In exceptional cases, this factor is cancelled by one in
the numerator and then the standard procedure (2.1) can be employed; we discuss
this special situation in 3.2.)

Because of the appearance of this factor, it is convenient to write

$
( , ) ( ) ( ) ( ) .... Y x x x x

= + + +
2 1 2
2
1
{ }


then the original equation, with y Y =
$
, gives

$
( )
$
( )
$
( )( ) + + =

Y p x Y q x Y x

2
1 2
2
,

so that possible solutions are
$
( , ) Y x
1
,
$
( , ) Y x
2
and
$
( , ) Y x

2
. But the second of
these, it turns out, is a multiple of the first (where there is a zero in the denominator of
the
i
s , i n > ); thus the second solution we require is
$
( , ) Y x

2
which follows the
same pattern as for repeated roots. We have avoided giving the details of this general
argument; we have simply aimed to indicate what we should expect. The precise
details, and any particular difficulties, will become evident when we tackle specific
problems of this type, as we shall now see.

Example 6

Use the method of Frobenius to find the general solution of the equation

xy y + = 0 ,

as a suitable power series about x = 0.

We see that there is a regular singular point at x = 0, so we may proceed by writing

y x a x
n
n
n
( ) =
+
=

0
( a
0
0 ),

and then we obtain

a n n x a x
n
n
n
n
n
n
( )( )

+ + + =
+
=

+
=

1 0
1
0 0
.

In the first series, we set n m = 1 , and in the second simply put n m = , to give

18
a m m x a x
m
m
m
m
m
m
+
+
=

+
=

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


i.e. a x a m m a x
m m
m
m
0
1
1
0
1 1 0

( ) ( )( ) + + + + + =

+
+
=

l q .

This is an identity for all x if

( ) = 1 0 and a m m a
m m +
+ + + + =
1
1 0 ( )( ) ( m= 0 1 2 , , , ... ).

So = 0 1 , they differ by an integer. If we select = 1 (the larger one), then the
recurrence relation becomes

( )( ) m m a a
m m
+ + + =
+
1 2 0
1
or a
a
m m
m
m
+
=
+ +
1
1 2 ( )( )
( m= 0 1 2 , , , ... ),

which is defined for all the given m. Thus

a
a
1
0
12
=
.
; a
a a
2
1 0
2 3 12 2 3
= =
. . . .
, etc.

to give a
m m
a
m
m
=

+
( )
( )( !)
1
1
2
0
, m= 1 2 3 , , , ... ,

and hence one solution is

y x a
x
n n
n n
n
( )
( )
( )( !)
=

+
+
=

0
1
2
0
1
1
.

Now if we select = 0 (the lower value), we obtain

m m a a
m m
( ) + =
+
1
1
( m= 0 1 2 , , , ... ),

so that, for m= 0, we get a
0
0 = , which is impossible; equivalently, we have division
by zero in the definition of the coefficient a
1
(exactly as we described in our
commentary earlier). The second solution, we believe, therefore contains a
logarithmic term, just as we encountered in Example 5. Thus we set

y x Y x x b x
n
n
n
( ) ( ) ln = +
=

0


where our first solution is written as y x a Y x ( ) ( ) =
0
; thus we obtain

19
x Y x Y
x
Y
x
b n n x Y x b x
n
n
n
n
n
n
+ +
F
H
G
I
K
J
+
L
N
M
M
O
Q
P
P
+ + =

ln . . ( ) ln
2 1
1 0
2
2
0 0
.

This is simplifies to give

m m b b x
x
Y Y
m m
m
m
( ) + + =
+
=

1
1
2
1
0
l q
= + = +
=

a x m a x m a x
m
m
m
m
m
m
m
m
m 0 0 0
2 1 1 2 ( ) ( ) ,

where a m m
m
m
= + ( ) ( )( !) 1 1
2
.

Thus the coefficients b
m
satisfy

m m b b m a
m m m
( ) ( ) + + = +
+
1 1 2
1
(for m= 0 1 2 , , , ... ),

so b
0
1 = , 2 3 3 2
2 1 1
b b a + = = (and we may choose b
1
0 = ), etc.; a second
solution is therefore

y x Y x x x x ( ) ( ) ln .... = + + + 1
3
4
2 3
76
4
e j
.

In conclusion, we have the complete, general solution as

y x A
x
n n
B x
x
n n
x x
n n
n
n n
n
( )
( )
( )( !)
ln
( )
( )( !)
... =

+
+

+
+ +
R
S
|
T
|
U
V
|
W
|
+
=
+
=

1
1
1
1
1
3
4
7
36
1
2
0
1
2
2 4
0
.


The exceptional case, where the roots for differ by an integer but no
logarithmic terms appear, will be discussed in detail when we consider the Bessel
equation (Chap. 3).

Comment: All our examples thus far have involved seeking power series about x = 0,
but this is not always what we require. Let us consider the equation

( ) 1 2 0 + + = x y xy y ;

this has a regular singular point at x = 1, so our conventional application of the
method of Frobenius is to seek a solution

y x a x
n
n
n
( ) ( ) =
+
=

1
0

.

20
However, it is often more convenient to transform the original equation so that the
standard power series, i.e. equivalently about x = 0, can be invoked. To do this, we
write
y x Y x Y z ( ) ( ) ( ) = = 1 where z x = 1,

and then we obtain
+ + + = zY z Y Y ( ) 1 2 0.

This equation has a regular singular point at z = 0 (i.e. x = 1), and so we may seek a
solution

Y z a z
n
n
n
( ) =
+
=

0
the standard form.

We conclude with one more example which contains two regular singular points, and
show what happens when we expand about each in turn. In this case we will write a
little more about the convergence of the series that we generate (and we will take the
opportunity to mention one of the special cases that can arise, but we will write more
of this later).

Example 7

Use the method of Frobenius to find a power-series solution, about x = 0, of the
equation
x x y xy x y
2 1
4
1 3 0 ( ) ( ) + = .

Examine the convergence of this solution, and then construct a corresponding solution
about the other singular point.


It is clear that we have a regular singular point at x = 0 (and also at x = 1), so we
write
y x a x
n
n
n
( ) =
+
=

0
,

to obtain

( ) ( )( ) ( ) ( ) x a n n x a n x x a x
n
n
n
n
n
n
n
n
n
+ + + + =
+
=

+
=

+
=

1 1
1
4
3 0
0 0 0




which is rewritten as
a m m x
m
m
m

+
=

+ + +
1
1
4
1
1 2 ( )( )

o t

+ + + + + =
+
=

a m m m x
m
m
m
( )( ) ( )

1 0
3
4
0
o t
.
21
Thus we have


a
0
3
4
1 + ( )
o t

+ + + + + + + =

+
=

a m m a m m x
m m
m
m
1
1
4
3
4
1
1 2 2 0 ( )( ) ( )( )

{ }
,

which is an identity for all x if


2 3
4
2 0 + = ( a
0
0 )

and a m m a m m
m m
( )( ) ( )( ) + + + = + + +

2 1 2
3
4
1
1
4
( m= 1 2 , , ... )

i.e. ( ) = 1 0
2 1
4
; a m m a m
m m
+ + = +

1
2
3
2
1
3
2
2
e je j e j
( m= 1 2 , , ... ).

So we have = 1 2 and = 3 2 (which differ by an integer); for = 3 2 we obtain

a m m a m
m m
( ) + =

1
1
2
or a
m
m
a
m m
=
+
F
H
G
I
K
J

1
1
( m= 1 2 , , ... ).

This gives one solution:

y x a x x x x
r
r
( ) .... .... = + + + + +
+
0
3 2 1
2
1
3
2 1
1
1
e j
.

On the other hand, for = 1 2 , the recurrence relation becomes

a m m a m
m m
( ) ( ) =

1 1
1
2
for m= 1 2 , , ... ,

which shows that, on m= 1, we have both a
0
and a
1
arbitrary. (Comment: This
special case of values of differing by an integer, but for which the recurrence is
valid even for the lower value of , is one that we shall discuss more fully later. The
procedure always involves the testing of the recurrence relation, and we use it if it
does not produce an inconsistent solution and, as we have seen, it is valid here.)
Thus we now have

a
m
m
a
m m
=

F
H
G
I
K
J

1
1
for m= 2 3 , , ... ,

which produces the solution

y x x a a x a x ( ) .... = + + +
1 2
0 1
1
2
1
2
e j


22
= + + + + + +
+
a x a x x x x
r
r
0
1 2
1
3 2 1
2
1
3
2 1
1
1 ... ...
e j
,

the complete, general solution. (Comment: We see that the solution obtained for
= 3 2 merely recovers one contribution to the general solution; it is always the case
that, if the recurrence gives a solution for both values of ,when they differ by an
integer, then the smaller value will always generate the complete, general solution.)

Now the infinite series

1
1
2
1
3
2 1
1
+ + + +
+
x x x
r
r
... ...

can be investigated, for example, by applying the ratio test; this shows us that the
series is certainly convergent if

x r
x r
r
r
( ) +
<

1
1
1
as r i.e. if x <1,

and it is divergent for x >1. For x = 1, we must examine the series directly: this
gives
1
1
2
1
3
1
4
+ + ....

and for the lower sign ( x = 1) this series converges, but it does not for the upper sign
( x = +1). Thus we have constructed a power-series representation of the solution
which is valid for < 1 1 x and the inadmissible choice x = +1 corresponds
precisely to the other singular point of the equation. In order to find the form of the
solution valid in the neighbourhood of x = 1, we now seek a power-series solution
about x = 1.

First, therefore, we set x z = 1 and write y x Y x Y z ( ) ( ) ( ) = = 1 , to give

( ) ( ) ( ) 1 1 2 0
2 1
4
+ + + = z zY z Y z Y ;

we seek a solution Y z a z
n
n
n
( ) =
+
=

0
(where we have elected to retain the standard
notation although, presumably, and a
n
will take different forms here). We obtain
( ) ( )( ) ( ) ( ) 1 1 1
2 1
0
1
0
+ + + + + +
+
=

+
=

z a n n z z a n z
n
n
n
n
n
n



=
+
=

1
4
0
2 0 ( ) z a z
n
n
n

,
which becomes

23
a m m m z
m
m
m
+
+
=

+ + + + + +
1
1
1 1 ( )( ) ( )

l q
+ + + + +
+
=

a m m m z
m
m
m
2 1
1
2
0
( )( ) ( )

o t

+ + + + =

+
=

a m m z
m
m
m
1
1
4
1
1 2 0 ( )( )

o t
.

With m= 1, we obtain
2
0 = , so = 0 is a repeated root (implying that
logarithmic terms will definitely arise); with m= 0, we get a a
1
1
2
0
0 = , leaving
(with = 0)

a m a m m a m m
m m m +
+ + + + =
1
2 2 1
2
1
1
4
1 2 1 2 0 ( ) ( )( )
e j
, m= 1 2 , , ... .

This generates the coefficients a a
2 0
8 = , a a
3 0
16 = , etc., so we have one solution

Y z a z z z ( ) ... = + + +
0
1
2
1
8
2 1
16
3
1
e j
.

A second solution necessarily takes the form

Y z Y z z b z
n
n
n
( )
$
( ) ln = +
=

0
,

where
$
( ) ... Y z z z = + + 1
1
2
1
8
2
; then we obtain

z z Y z
z
Y
z
Y b n n z
n
n
n
( )
$
ln .
$
.
$
( ) 1
2 1
1
2
2
2
2
+ + +
L
N
M
M
O
Q
P
P

+ + + +
L
N
M
M
O
Q
P
P
+
L
N
M
M
O
Q
P
P
=

+
=

( )
$
ln .
$
( )
$
ln 1
1 1
4
2 0
1
1 0
z Y z
z
Y b nz z Y z b z
n
n
n
n
n
n

.

But
$
Y is a solution of the original equation, so we are left with

( ) ( ) ( ) ( ) 1 1 1 2
2 1
2
1
1
1
4
0
+ + +

+
=

z b n n z z b nz z b z
n
n
n
n
n
n
n
n
n

= + + ( )
$
( )
$
1 2 1
2
z Y z Y
= + + + + + + ( ) ... ( ) ... 1 1 2 1
1
2
1
8
2 2 1
2
1
4
3
16
2
z z z z z z
e j e j
.

24
Then terms z
0
give b b
1
1
2
0
0 = , and we may select b
0
0 = , so that b
1
0 = . Terms z
1

give 4 0
2
b = , and terms z
2
then give 9 0
3
b = , and so on. There is a solution for
which b
n
= 0 for all n, so the complete, general solution can be written as

Y z A z z B z z z ( ) ... ... ln = + + + + + 1 1
1
2
1
8
2 1
2
1
8
2
e j e j


or y x x x x A B x ( ) ( ) ( ) ( ) ... ln = + + + + 1 1 1 1 1
1
2
1
8
2 1
16
3
c h
.

We conclude, therefore, that the singular point at x = 1 corresponds to a
logarithmic singularity in the solution here. Thus if we want to use the solution not
too near x = 1, we should employ the power-series representation about x = 0.
However, if we wished to capture more accurately the behaviour near x = 1 then we
should use the power series valid about x = 1 (i.e. about z = 0). (Note that the
solution does not exist on x = 1 unless B = 0.)

Comment: You might recognise 1
1
2
1
3
2 1
1
+ + + + +
+
x x x
r
r
... ... as the Maclaurin
expansion of ln( ) 1 x , valid for < 1 1 x , and also that 1
1
2
1
8
2 1
16
3
+ + + z z z ... is
the binomial expansion of ( ) 1
1 2
+ z . Indeed, our original equation has the exact
general solution
y x x A B x ( ) ln = + 1
c h
.


Exercises 2

1. Use the method of Frobenius to find the general solutions of these equations as
power series about x = 0. In each case, find explicitly the first three terms of each
series and also record the recurrence relation that will generate the rest.

(a) 4 2 0 xy y y + + = ; (b) xy x y y + + + = ( )
1
2
0;
(c) 2 7 1 3 0
2
x y x x y y + + = ( ) ; (d) x y x x y y
2
2 2 0 + + = ( ) ;
(e) xy y xy + + = 0; (f) x y x x y x x y
2 2
3 1 0 + + + + + = ( ) ( ) .

2. Repeat Q.1, but now find the power series about x = 1 for the equation

2 1 3 4 2 24 1 0
2
( ) ( ) x x x y y x y + =
e j
.

****************
**********



25
3. The Bessel equation and Bessel functions


An important equation in applied mathematics, engineering mathematics and most
branches of physics is the second order, linear, ordinary differential equation

x y xy x y
2 2 2
0 + + =
e j
,

where 0 is a real number; this is Bessels equation of order . [Friedrich W.
Bessel (1784-1846), German astronomer and mathematician; first to use the method
of parallax to determine the distance away of a star; developed his equation, and its
solution, in his study of the perturbations of planetary orbits.] We shall use the
method of Frobenius to construct solutions of this equation, for various ; we note
that this is permitted because the equation possesses a regular singular point at x = 0.

3.1 First solution

We set
y x a x
n
n
n
( ) =
+
=

0
( a
0
0 )

to give
a n n x a n x x a x
n
n
n
n
n
n
n
n
n
( )( ) ( )

+ + + + + =
+
=

+
=

+
=

1 0
0 0
2 2
0
e j
;

in the first, second and the constant multiple of the third series we set n m = , and in
the other series we write n m + = 2 . Thus we obtain

a x a x
0
2 2
1
2 2 1
1

+ +
+
e j
( )
+ + + =

a m a x
m m
m
m
( )
2 2
2
2
0
{ }
,

which is an identity for all x if

= ( a
0
0 ); a
1
0 = ( + 1 ) and a m a
m m
( ) + =

2 2
2
( m= 2 3 , , ... ).

For = + , we certainly have a solution with

a
a
m m
m
m
=
+
1
2 ( )
, m= 2 3 , , ... ,

and 0
1 3 5
= = = = a a a .... . Thus we obtain

26
a
a
2
0
2 2 1
=
+ . ( )
, a
a a
4
2 0
3
4 2 2
4 2 1 2
=
+
=
+ +
. ( )
. ( )( )

, etc.,

giving a
a
m m m
m
m
m
2
0
2
1
2 1 1
=

+ + +
( )
( !)( )( )...( )
,

which can be expressed in terms of the gamma function:

( )

=

z
t t
t 1
0
e d ( > 0 )

as a
m m
a
m
m
m
2
2
0
1 1
2 1
=
+
+ +
( ) ( )
( !) ( )

.

Thus we have a solution

y x a
m m
x
m
m
m
( ) ( )
( )
( !) ( )
= +

+ +
F
H
G
I
K
J
=
+
0
0
2
2 1
1
1 2

which is usually written as the Bessel function of the first kind, of order :

J

( )
( )
( !) ( )
x
x
m m
x
m m
m
=
F
H
G
I
K
J

+ +
F
H
G
I
K
J
=

2
1
1 2
2
0

.

This series converges for all x 0 . Examples of these Bessel functions are given in
the two figures below.

The Bessel functions J J J J
0 1 2 3
, , , , identified in this order by the ordering,
from left to right, of the first peak on each curve.

27

The Bessel functions J J J J
0 1 2 1 3 2
, , , , identified in this order by the ordering,
from left to right, of the first peak on each curve.

3.2 The second solution

From our work with the method of Frobenius, in the previous chapter, we know
that the second solution of these equations depends critically on the relationship
between the two roots for . First we note that only for = 0 are the roots repeated,
and this case is essentially the same as = n (n integer), for then will differ by an
integer; in these cases, we must expect that logarithmic terms will arise (unless
something exceptional occurs). In addition, there are non-integral values of which
also give rise to s that differ by an integer: all the half-integers e.g. = 1 2 , for then
= 1 2 (which differ by 1). For all other , the solution that we have just described
is appropriate when is replaced by , the solutions then being J

( ) x ; this is the
second solution.

In the case = n ( n = 1 2 , , ... ), with = = n , the recurrence relation becomes

a m m n a
m m
( ) =

2
2
( m= 2 3 , , ... )

which is not defined when m n = 2 , for a given n. Thus this case, and that of a
repeated root, will certainly require the inclusion of a logarithmic term. The
calculation follows that described in Chapter 2 and, although it is rather cumbersome
and lengthy, it is altogether routine. So, for example, for n = 0 , we find that the
second solution can be written

y x x x
x x
( ) ( ) ln
!
.... = +
F
H
G
I
K
J

F
H
G
I
K
J
F
H
G
I
K
J
+ J
0
2 2 4
2
3
2
1
2 2
.

However, this solution is usually combined with J
0
( ) x in the form

28
Y J
0 0
2 2 4
2
2 2
1
1
2
1
2 2
( ) ln ( )
!
x
x
x
x x
= +
F
H
G
I
K
J
F
H
G
I
K
J
R
S
T
+
F
H
G
I
K
J
+
F
H
G
I
K
J
F
H
G
I
K
J
F
H
G
I
K
J

+ + +
F
H
G
I
K
J
F
H
G
I
K
J
F
H
G
I
K
J

U
V
|
W
|
1
1
2
1
3
1
3 2
2 6
!
....
x
,
where =

z
e d
t
t t ln
0
0 5772 is Eulers constant; the resulting solution, Y
0
, is
the Bessel function of the second kind, of order zero. Correspondingly, with = n
( > 0 ), we find

Y J
n n
r n
r
n
x
x
x
n r
r
x
( ) ln ( )
( )!
!
= +
F
H
G
I
K
J
F
H
G
I
K
J


F
H
G
I
K
J
R
S
|
T
|

2
2
1
2
1
2
2
0
1

+ + + +
F
H
G
I
K
J
+
+ + + +
+
F
H
G
I
K
J
+
U
V
|
W
|
+
+
1
2
1
2
1
2
1 1
1 1 2
1
2
1
3
1 1
2
1
1
2
( ... )
!
( ) ( ... )
( !)[( )!]
....
n
n
n
n
n
x
n
x
.

Finally, we consider the case where is half-integer. The recurrence relation now
becomes, with = and = + ( ) 1 2 2 n ( n = 0 1 2 , , , ... ),

a m m n a
m m
=

1
2
2 e j
( m= 2 3 , , ... ),

and we see that m n
1
2
0 for every valid choice of m and n; thus the recurrence is
defined and the second solution exists without a logarithmic term and it simply
takes the form J

( ) x for the appropriate . Some examples of these Bessel

functions are shown in the next two figures; note that all these solutions diverge as
x 0 .


The Bessel functions Y , Y , Y , Y
0 1 2 3
, identified in this order by the ordering,
from left to right, of the first peak on each curve.
29


The Bessel functions J J J J
7 4 3 2 1 2 3 4
, , , , identified in this order by the ordering,
from left to right, of the first peak (ignoring the asymptote along the y-axis) on each curve.

Example 8

Write down the general solution of the equation xy y a xy + + =
2
0, where a > 0 is a
constant.

When the equation is multiplied by x it becomes almost a zero-order Bessel
equation; it can be recast precisely in this form if we write y x u x ( ) ( ) = , where is
a constant to be determined. The equation becomes

x u x u a x u
2 2 2 2
0 + + = ,

or ( ) ( ) ( )

x u x u
a
x u
2
2
2
0 + +
F
H
G
I
K
J
= ,

so we choose = a : z u zu z u
2 2
0 + + = ( u u z = ( ) , z x = ). This is the zero-order
Bessel equation, with general solution u z A z B z ( ) ( ) ( ) = + J Y
0 0
i.e.

y x A ax B ax ( ) ( ) ( ) = + J Y
0 0
.


3.3 The modified Bessel equation

We comment briefly on an equation that is intimately related to the conventional
Bessel equation, namely x y xy x y
2 2 2
0 + + =
e j
, called the modified Bessel
30
equation of order . We observe that the only change occurs in the term x y
2
it is
negative here. This immediately suggests that we should consider, formally at least, a
change of variable x z = i in the standard Bessel equation:

x y xy x y
2 2 2
0 + + =
e j
.

This yields z Y zY z Y
2 2 2
0 + + =
e j
, y z Y z ( ) ( ) i = ,

for which we have a solution. Thus, for example, we define

I i J i
n
n
n
r n
r
z z
z
r r n
( ) ( )
( )
!( )!
= =
+

+
=

2
2
0
,

which is the modified Bessel function of the first kind, of order n. The function that
corresponds to Y
n
is usually defined by

K i J i iY i
n
n
n n
z z z ( ) ( ) ( ) = +
+

2
1
,

which is the modified Bessel function of the second kind, of order n. Written
explicitly, for n = 0 (and reverting to the more conventional x), we obtain

I
0
2
2
4
2
6
2
1
2
0!)
2
2
2
3
( )
( )
(
( )
( !)
( )
( !)
.... x
x x x
= + + + +

K I
0 0
2
2
1
2
4
2
1
2
1
3
6
2
2 1
2
1
1
2
2
1
2
3
( ) ln( ) ( ) ( )
( )
( !)
( )
( !)
( )
( !)
.... x x x
x x x
= + + + + + + + + b g
e j e j
;

some examples of modified Bessel functions are shown in the figure below.


The modified Bessel functions K K I I
0
, , ,
1 1 0
, identified in this order by the ordering,
from left to right, of the upper part of the curves.
31
Exercises 3

1. Write the general solutions of these equations in terms of Bessel functions:

(a) x y xy x y
2 2
1 0 + + =
e j
; (b) 9 9 9 4 0
2 2
x y xy x y + + =
e j
.

2. Seek a solution of the equation xy y xy + + = ( ) 1 2 0 in the form
y x x Y x ( ) ( ) =

; use your result to find the general solution of the equation

xy y xy + = 2 0.

3. Show that the differential equation

d
dx
x y bx y
a c
+ =
e j
0 ( c a + 2 0, b > 0)

can be transformed into a Bessel equation for Y X ( ) :

X Y XY X Y
2 2 2
0 + + =
e j


by X
b
x =

, Y x y =

, for suitable and . Find and and hence obtain
the general solution of the equation

xy y y + + = 3 0.

4. In the definition of the Bessel function of the first kind ( J

), select = 1 2 and
hence show that

J
1 2
2
=
x
x sin , J

=
1 2
2
x
x cos .

5. The Bessel functions of the first kind, with integer orders, can be represented by

J d
n
x n x x ( ) cos sin =
z
1
0

b g .

Use this expression to show that

d
d
J J
x
x x x x
n
n
n
n
( ) ( ) =
1
;
d
d
J J
x
x x x x
n
n
n
n

+
= ( ) ( )
1
,

and hence obtain the recursion formula

32
J J J
n n n
x
n
x
x x
+
=
1 1
2
( ) ( ) ( ) .

5. Use the recursion formula in Q.5, and the results in Q.4, to obtain closed-form
representations of J
3 2
( ) x and J
3 2
( ) x .

****************
**********









































33
4. The Legendre polynomials

Another equation that plays a significant rle in many branches of the applications
of mathematics (particularly in problems that involve spherical symmetry) is the
Legendre equation

1 2 0
2
+ = x y xy y
e j
,

where is a parameter. [A.-M. Legendre (1752-1833), French mathematician who
made contributions to number theory, the theory of elliptic functions and to celestial
mechanics, which is where his equation first arose.] This equation has regular singular
points at x = 1, and so the domain of interest is normally < < 1 1 x (for general );
however, it is usual to seek a power-series solution about x = 0 (which is, of course,
an ordinary point). Thus we write

y x a x
n
n
n
( ) =
=

0


to give

1 1 2 0
2 2
0
1
0 0
+ =

x a n n x a nx a x
n
n
n
n
n
n
n
n
n
e j
( )

which becomes

a m m x a m m m x
m
m
m
m
m
m
+
=

+ + + =
2
0 0
2 1 1 2 0 ( )( ) ( ) l q .

This equation is an identity for all x if

a m m a m m
m m +
+ + = +
2
1 2 1 ( )( ) ( ) , m= 0 1 2 , , , ... ,

with both a
0
and a
1
arbitrary. Thus, for general , we have a general solution:

y x a x x x ( ) ( ) ( )( ) ... = + +
0
1
2
2 1
24
4 1
720
6
1 6 6 20
+ + + a x x x
1
1
6
3 1
120
5
2 2 12 ( ) ( )( ) ... .

The ratio test then gives

a x
a x
m
m
m
m
+
+
<
2
2
1 as m i.e.
m m
m m
x
( )
( )( )
+
+ +
<
1
1 2
1
2

as m ,

so that each series that contributes to the general solution converges for < < 1 1 x .
34

In many applications we need bounded solutions for 1 1 x and, for general
, this is impossible: the series are not convergent at x = 1. (Indeed, a Frobenius
approach about x = +1 or x = 1 shows that the solution diverges like ln( ) x m1 as
x 1.) However, if the series possesses a finite number of terms i.e. it is
polynomial, then it certainly will remain bounded for all x [ , ] 1 1 ; we then say that
the series terminates. This situation will occur here for any = + n n ( ) 1 ,
n = 0 1 2 , , , ... ; the first few such solutions are then

n = 0 , y = 1; n = 1, y x = ; n = 2 , y x = 1 3
2
; n = 3, y x x =
5
3
3
,

and so on. Of course, these are still solutions when multiplied by constants, and when
these are chosen to ensure that y( ) 1 1 = , for each n, they are known as the Legendre
polynomials ( P
n
x ( ) ), so we have

P
0
1 ( ) x = , P
1
( ) x x = , P
2
1
2
2
3 1 ( ) x x =
e j
, P x x x
3
1
2
3
5 3 ( ) =
e j
, etc. .

These first few polynomials are shown in the figure below.




The Legendre polynomials, P P P P
0
, , ,
1 3 2
, identified in this order by
the number of times (n) that each crosses the x-axis.

Exercises 4

1. Obtain the Legendre polynomials P
4
( ) x and P
5
( ) x .

2. Confirm that P
d
d
n
n
n
n
n
x
n x
x ( )
!
=
L
N
M
O
Q
P
1
2
1
2
e j
for n = 0 1 2 3 4 , , , , . (This is called
Rodrigues formula.)

35
3. Legendres equation in the form 1 2 1 0
2
+ + = x y xy n n y
e j
( ) has one solution
y x x
n
( ) ( ) = P ; use the method of variation of parameters to find a second solution
(usually written Q
n
x ( ) ) in the cases n = 0 and n = 1.

4. Use Rodrigues formula (Q.2) to obtain the recursion formula

( ) ( ) ( ) ( ) ( ) n x n x x n x
n n n
+ = +
+
1 2 1
1 1
P P P .

****************
**********

































36
5. The Hermite polynomials

Another standard second order ODE is

+ = y xy y 2 0 , < < x ,

which appears, most typically, in elementary solutions of Schrdingers equation;
is a parameter. This is Hermites equation, where special choices of give rise to the
Hermite polynomials. [Charles Hermite (1822-1901), French mathematician who
made important contributions to the theory of differential equations and also worked
on the theory of matrices; in 1873 he proved that e is a transcendental number.] The
equation has no singular points all (finite) points are ordinary points so we seek a
solution
y x a x
n
n
n
( ) =
=

0


which gives
a n n x a nx a x
n
n
n
n
n
n
n
n n
( ) + =

1 2 0
2
0 0 0
.

This is written as

a m m x a m x
m
m
m
m
m
m
+
=

+ + =
2
0 0
2 1 2 0 ( )( ) ( )

which is an identity for all values of x if

a m m a m
m m +
+ + =
2
2 1 2 ( )( ) ( ) , m= 0 1 2 , , , ... ,

with both a
0
and a
1
arbitrary. This recurrence relation gives directly the general
solution

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

It is immediately clear that there exists a polynomial solution of the original equation
whenever = 2n , n = 0 1 2 , , , ... . With the choice = 2n , and the arbitrary
multiplicative constant chosen so that the coefficient of the term x
n
is 2
n
, the
resulting solution is the Hermite polynomial, H
n
x ( ) . Thus we have

H
0
1 ( ) x = , H
1
2 ( ) x x = , H
2
2
4 2 ( ) x x = , H
3
3
8 12 ( ) x x x = ,
H
4
4 2
16 48 12 ( ) x x x = + , etc.;

37
the first four of these are shown in the figure below.



The Hermite polynomials, H H H H
0
, , ,
1 3 2
, identified in this order by
counting up on the far right of the figure.


Exercises 5

1. Write down H
5
( ) x .

2. Confirm that H e
d
d
e
n
n x
n
n
x
x
x
( ) ( ) =
F
H
I
K

1
2 2
for n = 0 1 2 3 4 , , , , . (This is
Rodrigues formula for the Hermite polynomials.)

3. The equation + = y xy ny 2 2 0 has one solution y x
n
= H ( ) ; use the method of
variation of parameters to find a second solution in the cases n = 0 and n = 1.

4. Use Rodrigues formula (Q.2) to obtain the recursion formula

H H H
n n n
x x x n x
+
=
1 1
2 2 ( ) ( ) ( ) .

****************
**********












38
6. Generating functions

We conclude with an additional and intriguing property of the functions that we
have discussed in this Notebook (and similar properties arise for other functions and
polynomials). The existence of recursion formulae, and also the Rodrigues formulae
[O. Rodrigues (1794-1851), French mathematician] for the polynomials, is related to
another important idea: the generating function.

Suppose that we are given a set of functions, F x
n
( ) say; further, suppose that these
can be obtained as the coefficients of some power series (in z, say) i.e. we write

F x z
n
n
n
( )
=

0
.

If we can find a function, f z x ( ; ) , whose Maclaurin expansion in z recovers the
series, so that we have
f z x F x z
n
n
n
( ; ) ( ) =
=

0
,

we call f z x ( ; ) a generating function. (It obviously generates, in a purely algebraic
way, the set of functions that are of interest.) An elementary example of this idea is
the choice
f z x z x ( ; ) = +

which has the Maclaurin actually binomial in this case expansion

f z x x
z
x
z
x
z
x
( ; ) .... = +
F
H
G
I
K
J
+
F
H
G
I
K
J

R
S
|
T
|
U
V
|
W
|
1
1
2
1
8
1
16
2 3
;

thus we have generated the set of functions

x
x x x
x x
, , , , ....
1
2
1 1
8
1 1
16
1
2


(and the coefficients that appear here may, or may not, be included). We may
therefore regard f z x z x ( ; ) = + as a generating function for F x x
n
n
( ) =

1
2
.

For each of our three examples (Bessel, Legendre, Hermite), simple generating
functions exist; we will present them detailed proofs of their correctness are beyond
this text and make a few observations about each.

6.1 Legendre polynomials

First we consider the function 1 1 2
2
+ xz z , which has the expansion (in z)
39
1 2 1 2 2
2
1 2
1
2
2 3
8
2
2
+ = + + +

xz z xz z xz z
e j e j e j
....

= + + + 1
3
2
2 1
2
2
xz x z
e j
....

and we recognise the coefficients of z
0
, z
1
, z
2
: 1, x ,
1
2
2
3 1 x
e j
, respectively.
These are the first three Legendre polynomials; see Chapter 4. In general, we find that

1
1 2
2
0 +
=
=

xz z
x z
n
n
n
P ( )

and this property can be analysed in a number of ways; the most powerful is to
differentiate each side with respect to z and, separately, with respect to x. First, the z-
derivative gives

+ + =

1
2
2
3 2
1
0
1 2 2 2 xz z x z x nz
n
n
n
e j
b g P ( )

which can be written (upon using the original definition of the generating function) as

( ) ( ) ( ) x z x z xz z n x z
n
n
n
n
n
n
= +
=

P P
0
2 1
0
1 2
e j


or +

P P
m
m
m
m
m
m
z x z
1
1 0

= + +
+
=

( ) ( ) 1 2 1
1
1 0
1
1
m z x m z m z
m
m
m
m
m
m
m
m
m
P P P .

This is an identity in z if

xP P
0 1
= and x m xm m
m m m m m
P P P P P = + +
+ 1 1 1
1 2 1 ( ) ( ) ( m= 1 2 , , ... )

and the first of these is automatically satisfied; the second can be written as

( ) ( ) ( ) ( ) ( ) n x x n x n x
n n n
+ = +
+
1 2 1
1 1
P P P ,

which is the recursion formula for Legendre polynomials (see Exercises 4).

On the other hand, if we differentiate with respect to x, we obtain

+ =

1
2
2
3 2
0
1 2 2 xz z z x z
n
n
n
e j
( ) ( ) P
40
or z x z xz z x z
n
n
n
n
n n
P P ( ) ( ) = +
=

1 2
2
0 0
e j
,

which can be simplified when we use the previous derivative:

z n x z x z x z
n
n
n
n
n
n
P P ( ) ( ) ( )

=
1
0 0
.

This is conveniently written as

m x z x x z x z
m
m
m
m
m
m
m
m
m
P P P ( ) ( ) ( )
=

=
0 0
1
1


which is an identity in z if

= P
0
0 and m x
m m m
P P P =
1
( m= 1 2 , , ... );

we know that = P
0
0 , so we are left with

=

P P P
n n n
x x x n x
1
( ) ( ) ( )

another type of recurrence relation.

6.2 Hermite polynomials

The function that we consider here is

exp 2
2
xz z
e j


which possesses the expansion

exp
( )
!
( )
!
( )
!
...
!
... 2 1 2
2
2
2
3
2
4
1
2
2
2 3 4
2
4
xz z xz
xz xz xz
z
z
= + + + + +
F
H
G
I
K
J
+ +
F
H
G
I
K
J
e j

= + + + + 1 2 2 1 2
2 2 4
3
3 3
xz x z x x z
e j e j
...

and again we recognise (see Chapter 5) the first four H s
n
, in the form

H H H H
0 1
1
2
2
1
6
3
, , , .

Thus we have a generating function for the Hermite polynomials, written as

41
exp ( )
!
2
2 1
0
xz z x z
n
n
n
n
=
=

e j
H

(with the usual identification: 0! 1 = ). This generating function can be used (as
described in the preceding section) to derive various recurrence relations.

The z-derivative yields

2 2
2 1
1
1
1
( ) exp ( )
( )!
x z xz z x z
n
n
n
n
=


=

e j
H

or 2
1
0
1
1
1
1
( ) ( ) ( )
! ( )!
x z x z x z
n
n
n
n
n
n
n
n
=
=

H H

which can be written as

2 2
1
0
1
1
1
1
1
1
0
x x z x z x z
m
m
m
m
m
m
m
m
m
m
m
m
! ( )! !
( ) ( ) ( ) H H H
=

+
=

= .

This is an identity in z if

2
0 1
xH H = and
2 2
1
1
1 1
x
m m m
m m m
! ( )! !
H H H

=
+
( m= 1 2 , , ... );

the first of these is automatically satisfied, and the second becomes

H H H
n n n
x x x n x
+
=
1 1
2 2 ( ) ( ) ( ) ( n = 1 2 , , ... ),

which appears in Exercises 5.

6.3 Bessel functions

Finally, we return to the functions which possess a far more complicated structure:
the Bessel functions (discussed in Chapter 3). We shall present the relevant details for
the Bessel functions J
n
x ( ) (for n = 0 1 2 , , , ... ) by considering the function

exp
1
2
1
z
z
x
F
H
G
I
K
J
R
S
T
U
V
W
( z 0)

which, we observe, will necessarily generate a more complicated powers series (in z)
than has arisen in our earlier examples. This time we must expand in a Laurent series
(i.e. terms z
n
and z
n
will appear), producing a power series that is convergent for
0 < < z . The expansion is

42
e e
xz x z
xz xz xz
2 2
2 3
1
2
1
2
2
2
3
.
( )
!
( )
!
( )
!
....

= + + + +
F
H
G
I
K
J

+ +
F
H
G
I
K
J
1
2
1
2
2
2
3
2 3
( )
!
( )
!
( )
!
....
x z x z x z

=
F
H
G
I
K
J
+
F
H
G
I
K
J

R
S
|
T
|
U
V
|
W
|
+
F
H
G
I
K
J
+
F
H
G
I
K
J

R
S
|
T
|
U
V
|
W
|
1
2
1
2
2 2
1
1
2 2
1
2 3 2
2
2
4 2 4
x x
z
x x x
( !)
...
! !
...

F
H
G
I
K
J
+
F
H
G
I
K
J

R
S
|
T
|
U
V
|
W
|
+
1
2
1
1
2 2
1
2 3 2
2 4
z
x x x
! !
.... .....
=
=

z x
n
n
n
J ( ) .

Thus we have demonstrated (in outline, at least) that

exp ( )
1
2
1
z z x z x
n
n
n
=

e j { }
J

which, as developed earlier, can be used to obtain various recurrence formulae. The z-
derivative gives

1
2
2 1
2
1 1
1+ =

=

z x z z x nz x
n
n
n
e j e j { }
exp ( ) J

or
1
2
1
z z x z x nz x
n
n
n
n
n n
+ =

e j
J J ( ) ( )

which leads directly to

J J J
n n n
x
n
x
x x
+
=
1 1
2
( ) ( ) ( ) (see Exercises 3).

Correspondingly, the x-derivative produces

1
2
1 1
2
1
z z z z x z x
n
n
n
=

=

e j e j { }
exp ( ) J

or
1
2
1
z z z x z x
n
n
n
n
n n
=

e j
J J ( ) ( )

which produces

43
=
+
J J J
n n n
x x x ( ) ( ) ( )
1
2
1 1
.

Exercises 6

1. Use the appropriate generating functions to find P
3
( ) x and H
4
( ) x .

2. Use the appropriate recursion formulae (and any results already available) to find
P
6
( ) x and H
6
( ) x .

3. Use the recursion formulae (obtained from the generating functions; see 6.1) to
show that
1 2 1 0
2
+ + = x x n n
n n n e j
P P P ( )

(thereby confirming that the generating function does produce functions P
n
).

4. Repeat Q.3 for H
n
x ( ) (6.2) and the equation + = H H H
n n n
x n 2 2 0.

5. Repeat Q.3 for J
n
x ( ) (6.3) and the equation + + =

J J J
n n n
x n x
1 2 2
1 0
e j
.

6. Use the generating function for the Bessel functions, J
n
, to show that

J J
n
n
n
x x ( ) ( ) ( ) = 1 and J J

=
n
n
n
x x ( ) ( ) ( ) 1 ,

and hence deduce that = J J
0 1
( ) ( ) x x and then that

J
d
d
J
n
n n
n
x x
x x
x ( ) ( ) ( ) =
F
H
G
I
K
J
1
1
0
.

****************
**********













44
Answers

Exercises 1

1. (a) Regular singular points at x = 0, x = 1; (b) regular singular point at x = 1 and
an essential singular point at x = 1; (c) no singular points all (finite) points are
ordinary points, but the point at infinity is an essential singular point.

Exercises 2

Where the result is easily obtained, the full series is presented.

1. (a) y x A
x
n
B x
x
n
n
n
n
n
( )
( )
( )!
( )
( )!
=

+

+
=

2 2 1
0 0
(with = 0 1 2 , );
(b) y x A x
x
n
Bx
n
n
x
n
n
n
n
( )
( )
!
( )!
!
( ) =

+
+

=

0 0
2 1
4 (with = 1 2 1 , ) and note that the
first solution here can be written A x
x
e

;

(c) y x A x x x Bx x x x ( ) ...
. .
= + + + +

1 1
7.1
2.9
13 7.7
2.4.9.11
2 3 21
5
49
5
2 343
15
3
e j e j
(with
= 3 1 2 , );

(d) y x Ax
x
n
B y x x x x x
n
n
( )
( )
!
( ) ln ... =

+ + +
=

2
0
1
2 3
4
4
{ }
(with = 1 2 , ) and note
that the first solution here can be written Ax
x 2
e

;

(e) y x A
x
n
B y x x x x
n
n
( )
( !)
( ) ln ... =

+ + +
=

2
2
0
1
1
4
2 3
128
4
4
e j
{ }
(with = 0 0 , );

(f) y x Ax x B y x x x ( ) ... ( ) ln ... = + + + +
1 1
4
2
1
1
2
1 1
e j o t
(with = 1 1 , ).

2. Set x z = 1 : 4 2 0 zY Y Y + = , then
y x A
x
n
B x
x
n
n
n
n
n
( )
( )
( )!
( )
( )!
=

+

=

1
2
1
1
2
0 0
(with = 0 1 2 , ) which can be written
y x A B x x ( ) cosh( ) = + 1 1
d i
.

Exercises 3

1. (a) y x A x B x ( ) ( ) ( ) = + J Y
1 1
; (b) y x A x B x ( ) ( ) ( ) = +

J J
2 3 2 3
.

2. x Y xY x Y
2 2 2
0 + + =
e j
; y x x A x B x ( ) ( ) ( ) = +

3 2
3 2 3 2
J J
e j
.
45
3. = +
1
2
2 ( ) c a , = + ( ) ( ) 1 2 a c a ; y x
x
A x B x ( ) ( ) ( ) = +
1
2 2
J Y b g .

6. J
3 2
2 1
( ) sin cos x
x x
x x =
F
H
G
I
K
J

; J

= +
F
H
G
I
K
J
3 2
2 1
( ) sin cos x
x
x
x
x

.

Exercises 4

1. P
4
1
8
4 2
35 30 3 ( ) x x x = +
e j
; P
5
1
8
5 3
63 70 15 ( ) x x x x = +
e j
.

3. Q
0
1
1
( ) ln x A B
x
x
= +
+

; Q
1
1
1
2 ( ) ln x Ax B x
x
x
= +
+

F
H
G
I
K
J
.

Exercises 5

1. H
5
5 3
32 160 120 ( ) x x x x = + .

3. n = 0 : y x x
x
x
( ) =
z
e d
2
; n = 1: y x x
x
x
x
x
( ) =
z
e
d
2
2
.

Exercises 6

1. P
3
1
2
3
5 3 ( ) x x x =
e j
; H
4
4 2
16 48 12 ( ) x x x = + .

2. P
6
1
16
6 4 2
231 315 105 5 ( ) x x x x = +
e j
; H
6
6 4 2
64 480 720 120 ( ) x x x x = + .

****************
**********
















46
Index
Bessel equation 25
modified 29
first kind 26
generating function 41
second kind 28
Bessel, F.W. 25
classification singular point 8
closed form 5, 7
convergent 5
ratio test 22
equation Bessel 25
Hermite 36
indicial 13
Legendre 33
modified Bessel 29
essential singular point 9
Euler's constant 28
Frobenius index 11
method of 11
power series 11
Frobenius, G.F. 11
function Bessel, 1st kind 26
Bessel, 2nd kind 28
gamma 26
generating 38
mod Bessel 1st kind 30
mod Bessel 2nd kind 30
gamma function 26
generating function 38
Bessel function 41
Hermite polynomials 40
Legendre polynomials 38
Hermite equation 36
Hermite polynomials 36
generating function 40
recursion formula 37
recursion formula 41
Rodrigues' formula 37
Hermite, C. 36
identity in x 6
index Frobenius method 11
roots (general) 13
roots differ by N 16
roots repeated 14
indicial equation 13
roots (general) 13
roots differ by N 16
roots repeated 14
infinity point at 10
introduction power series 5



47
Legendre equation 33
Legendre polynomials 34
generating function 38
recursion formula 35
Rodrigues' formula 34
Legendre, A.-M. 33
method of Frobenius 11
modified Bessel equation 29
modified Bessel function first kind 30
second kind 30
ordinary point 9
point at infinity 10
polynomials Hermite 36
Legendre 34
power series about x = a (not 0) 19
Frobenius 11
introduction 5
ratio test for convergence 22
recurrence relation 7, 13
recursion formula Hermite polynomials 37
Hermite polynomials 41
Legendre polynomials 35
regular singular point 8, 9
Rodrigues' formula Hermite polynomials 37
Legendre polynomials 34
Rodrigues, O. 38
series termination 34
singular point 9
classification 8
essential 9
regular 8, 9
solution closed form 5
termination of series 34