Special Functions

Chapter 5
SPECIAL FUNCTIONS
Chapter 5 SPECIAL FUNCTIONS
table of content
Chapter 5
Special Functions
5.1 Heaviside step function

- filter function
5.2 Dirac delta function
- modeling of impulse processes
5.3 Sine integral function
- Gibbs phenomena
5.4 Error function
5.5 Gamma function
5.6 Bessel functions
Bessel equation of order (BE)
Singular points. Frobenius method
Indicial equation
First solution Bessel function of the 1st kind
Second solution Bessel function of the 2nd kind.
General solution of Bessel equation
6. Bessel functions of half orders spherical Bessel functions
7. Bessel function of the complex variable Bessel function of the 3rd
kind (Hankel functions)
8. Properties of Bessel functions:
- oscillations
- identities
- differentiation
- integration
- addition theorem
9. Generating functions
10. Modified Bessel equation (MBE)
- modified Bessel functions of the 1st and the 2nd kind
11. Equations solvable in terms of Bessel functions
- Airy equation, Airy functions
12. Orthogonality of Bessel functions
- self-adjoint form of Bessel equation
- orthogonal sets in circular domain
- orthogonal sets in annular fomain
- Fourier-Bessel series
1.
2.
3.
4.
5.
5.7 Legendre Functions

1.
2.
3.
4.
5.
6.
Legendre equation
Solution of Legendre equation Legendre polynomials
Recurrence and Rodrigues formulae
Orthogonality of Legendre polynomials
Fourier-Legendre series
Integral transform
5.8 Exercises
Chapter 5
SPECIAL FUNCTIONS
Introduction
In this chapter we summarize information about several functions
which are widely used for mathematical modeling in engineering.
Some of them play a supplemental role, while the others, such as the
Bessel and Legendre functions, are of primary importance. These
functions appear as solutions of boundary value problems in physics and
engineering.
The survey of special functions presented here is not complete we
focus only on functions which are needed in this class. We study how
these functions are defined, their main properties and some
applications.
5.1 Heaviside Step Function
5.1 Heaviside Function (unit step function)

The Heaviside step function has only two values: 0 and 1 with a
jump at x = 0 :
0 x < 0
H (x ) =
1 x > 0
(1)
Graphically it can be shown as:

> plot(Heaviside(x),x=-3..3,discont=true);
H ( x)
Shifting of the step function along the x-axis:

0 x < a
H (x a ) =
1 x < a
(2)
> plot(Heaviside(x-2),x=-1..4,discont=true);
H ( x 2)
filter function
The filter function can be constructed in terms of the step function:

0 x<a
F (x , a , b ) = H (x a ) H (x b ) = 1 a < x < b
0 x>b
It cuts the values of functions to zero outside of the interval [a , b] :

> F(x,1,3):=Heaviside(x-1)-Heaviside(x-3);
> plot(g(x)*F(x,1,3),x=-1..5,discont=true);
g ( x ) F ( x,1,3 )
The Heaviside step function is used for the modeling of a sudden

increase of some quantity in the system (for example, a unit voltage is
suddenly introduced into an electric circuit) we call this sudden
increase a spontaneous source.
(3)
5.2 Dirac Function (delta function)

The Dirac delta function (x ) is not a function in the traditional sense it
is rather a distribution a linear operator defined by two properties. The
first describes its values to be zero everywhere except at x = 0 where the
value is infinite:
x =0
x0
(x ) =
(4)
The second property provides the unit area under the graph of the delta
function:
b
(x )dx = 1
where a < 0 and b > 0
The delta function is vanishingly narrow at x = 0 but nevertheless

encloses a finite area. It is also known as the unit impulse function.
The Dirac delta function can be treated as the limit of the sequence of
the following functions:
a) rectangular functions:
( x ) = lim Sh ( x ) = lim
h 0
H ( x + h) H ( x h)
2h
h 0
b) Gauss distribution functions:
( x ) = lim G ( x ) = lim
0
x2
c) triangle functions:
0
x 1
+
2 h
h ( x) = h
x +1
h2 h
(x ) = lim h (x )
h0
d) Cauchy density (distribution) functions:
( x ) = lim Dn = lim
n
e) sine functions:
(x ) = lim
sin nx
x
(1 + n2 x 2 )
x < h
h x < 0
0xh
x>h
Properties
1) Extension of the interval of integration to all real numbers still keeps

the unit area under the graph of the delta function:
(x )dx = 1
2) The Dirac delta function is a generalized derivative of the Heaviside

step function:
dH (x )
(x ) =
dx
It can be obtained from the consideration of the integral from the
definition of the delta function with variable upper limit. It is obvious,
that
x
0 x < 0
(t )dt = 1 x > 0 = H (x )
Therefore, the step function is formally an antiderivative of the delta

function which now can be interpreted as a derivative of a discontinuous
function.
3) Shifting in x :
(x a ) =
a +c
x=a
xa
(x a )dx = 1 ,
c >0
a c
4) Symmetry:
(x) = ( x)
( x a) = (a x)
5) Derivatives:
1
x
The derivative can be defined as a limit of triangle functions and
interpreted as a pure torque in mechanics. The higher order derivatives of
the delta function are:
k!
(k ) (x ) = ( 1)k k (x )
k = 1,2,...
x
(x ) = (x )
6) Scaling:
(ax ) =
1
(x )
a
for a 0
7) There are some important properties of the delta function which reflect
its application to other functions. If f (x ) is continuous at x = a , then
f (x ) (x a ) = f (a ) (x a )
c
f (x ) (x a )dx = f (a )
b<a<c
f (x ) (x a )dx = f (a )
f (t ) (t a )dt = f (a )H (x a )
x0
x0 a
Integration with derivatives of the delta function (integration by parts):
f ( x ) ( x )dx = f ( x ) ( x )
f ( x ) ( x )dx = f ( x ) ( x )
f ( x ) ( x )dx = f (0 )
f ( x ) ( x )dx = f (0 ) =
f ( 0 )
8) Laplace transform:
L{ (x )} = e sx (x )dx = 1
0
L{ (x a )} = e sx (x a )dx = e as
a >0
9) Fourier transform:
F { (x a )} = e ix (x a )dx = e ia
a >0
Applications
The delta function is applied for modeling of impulse processes.

For example, the unit volumetric heat source applied instantaneously at
time t = 0 is described in the Heat Equation by the delta function:
u
k 2 u = (t )
t
If the unit impulse source is located at the point r = r0 and releases all
energy instantaneously at time t = t0 , then the Heat Equation has a source
u
k 2 u = (t t0 ) (r r0 )
t
Impulse models are used for calculation of the Greens function for nonhomogeneous DE.
The other interpretation of the delta function (t t0 ) as a force applied
instantaneously at time t = t0 yielding an impulse of unit magnitude.
Example Consider IVP: unit impulse is imposed on a dynamical system
initially at rest at t = 5 :
y + 9 y = (t 5 )
initial conditions: y (0 ) = 0
y (0 ) = 0
Solution: Apply the Laplace transform to the given initial value problem
(use the property of the Laplace transform):
s 2Y + 9Y = e 5 s
Solve the algebraic equation for Y :
e 5 s
Y= 2
s +9
The inverse Laplace transform yields a solution of IVP:
1
y (t ) = H (t 5 ) sin 3(t 5 )
3
The graph of the solution shows that the system was at rest until the time
t = 5 , when an impulse force was applied yielding periodic oscillations.
5.3 Sine Integral Function

The sine integral function is defined by the formula:
x
Si( x ) =
0
sin t
dt
t
x ( , )
(5)
The integrand can be expanded in Taylor series and then integrated term
by term yielding a series representation of the sine integral function:
( 1)n x 2 n+1
n=0 (2 n + 1)(2 n + 1)!
Si (x ) =
(6)
Graphically it can be shown as:

> plot(Si(x),x=-25..25);
Si ( x )
The limiting values of the sine integral function are determined by the
Dirichlet integral
sin
d = 2
0
which was obtained in Section 3.5 (Remark 3.7, p.227) as a particular case
of the Fourier transformation of the step function.
In Chapter 3, we discussed connection of Gibbs phenomena in the Fourier

series approximations of functions with jumps and the properties of sine
integral function.
5.4 Error Function

The error function is the integral of the Gauss density function
2
erf ( x ) =
t 2
x ( , )
dt
(7)
> plot(erf(x),x=-4..4);
erf ( x )
The complimentary error function is defined as
erfc( x ) = 1 erf (x ) =
> plot(erfc(x),x=-4..4);
erfc( x )
Series expansion of the error function:
erf (x ) =
( 1)n x 2 n+1
n! (2n + 1)
n=0
Derivatives of the error function:

d
2 x2
erf (x ) =
e
dx
d2
4 x x2
erf (x ) =
e
2
dx
e
x
t 2
dt
(8)
5.5 Gamma Function

Definition
The Gamma function appears in many integral or series representations

of special functions. Gamma was introduced by Leonard Euler in 1729
who investigated the integral function
1
q
p
x (1 x ) dx
p,q
which for natural values p,q
is equal to
p! q!
( p + q + 1) !
With some transformation of this integral and taking the limits, Euler
ended up with the result
1
( ln x )
dx = n! ( n + 1)
Later, the gamma function was defined by the improper integral which
converges for all x except of 0 and negative integers (Euler, 1781):
(x ) = e t t x1dt
(9)
> plot(GAMMA(x),x=-5..4);
( x)
Properties:
a)
(x + 1) = x (x )
(x + 1)
(10)
= e t t ( x+1)1dt
0
= e t t x dt
0
= t x de t
0
= t x e t
+ e t dt x ( lim t x e t = 0 )
= x e t t x1dt
0
= x (x )
b)
When x = n is a natural number then
(n ) = (n 1)!
n = 1,2 ,3 ,...
(n + 1) = n!
n = 0 ,1,2 ,... provided that 0! = 1
(11)
The gamma function is a generalization to real numbers of a

factorial (which is defined only for non-negative integers).
Proof:
(1) = 1 then using property (a)

(2 ) = (1 + 1) = 1 (1) = 1 = 1!
(3) = (2 + 1) = 2 (2 ) = 2 1 = 2!
(then by mathematical induction)
c)
The gamma function does not exist at zero and negative integers.
d)
The gamma function is differentiable everywhere except at

x = 0 ,1,2 ,... . It is a differentiable extension of the factorial.
The derivative of the gamma function is called the digamma
function. It is denoted by
( x) = ( x)
( x)
e)
Stirling formula (approximation for large x , x > 9 )

x
e
(x + 1) 2x
f)
Calculation of gamma function: Lanczos approximation in

Fortran or C++ Numerical recipes.
g)
Binomial coefficients:
( z + 1)
z
z!
=
=
w w! ( z w ) ! ( w + 1) ( z w + 1)
(12)
5.6
Bessel Functions
1. Bessel Equation
In the method of separation of variables applied to a PDE in cylindrical

coordinates, the equation of the following form appears:
r 2 R(r ) + rR(r ) + 2 r 2 2 R(r ) = 0
r [0 , )
This equation is the second order linear ordinary differential equation with
variable coefficients. It includes two parameters and . It is not of
the Euler-Cauchy type. Can be solved by the Frobenius method.
Simplify equation by the change of variables to eliminate parameter :

y (x ) = R(r )
x = r
dR dy dx
=
= y
dr dx dr
d 2 R d dR d dy d dy dx
=
= 2 y

=
=
dr dr dr dx dx dx dr
dr 2
Then the equation becomes
x 2 y + xy + x 2 2 y = 0
(14)
Now the equation is written in traditional variables, and it includes only

one parameter . This equation is called to be a Bessel Equation of
order . Apply a power-series solution to this equation.
2. Singular Points
x2 = 0
x=0
Singular points of the differential equation with variable coefficients are

the points x at which the first coefficient becomes zero:
is the only singular point of the Bessel Equation. Therefore, if we find a
power-series solution around this point, it will be convergent for all real
numbers.
Determine the type of the singular point.
Divide the equation by x2 to rewrite it in the normal form:
2
1
y + y + 1 2 y = 0
x
x
Identify coefficients of the equation in normal form:

1
2
P(x ) = and Q(x ) = 1 2 .
x
x
Check if the singularity is removable:
xP(x ) = 1
is analytic: p0 = 1
x 2 Q(x ) = x 2 2 is analytic: q0 = 2
Therefore, x = 0 is a regular singular point, and the Frobenius theorem
can be used for solution of the Bessel Equation.
3. Indicial Equation
Substitute coefficients p0 and q0 into the indicial equation (Ch. 2.42):

r 2 + ( p0 1)r + q0 = 0
r 2 2 = 0
There are two roots of this equation:
r1 =
r2 =
(choose 0 for convenience, later we can abandon this assumption).
The Frobenius approach depends on the form of the difference of roots of
the indicial equation:
r1 r2 = ( ) = 2
Two cases of the Frobenius theorem may be involved:
a)
b)
r1 r2 = 2 integer
r1 r2 = 2 = integer
this case includes = n , n N 0 (positive integers and zero) and
2n + 1
=
(half of the odd integer)
2
In both cases, the first solution, following the Frobenius theorem, has to
be found in the form:
y1 = cn x n+r = cn x n+ , x > 0
(15)
n =0
n =0
Proceed to this solution, and then we will analyze how it handles the
abovementioned cases.
Using assumed form of solution (2), calculate the derivatives
4. First Solution
y1 = cn x n+
n =0
y1 = (n + )cn x n+ 1
n =0
y1 = (n + )(n + 1)cn x n+ 2
n =0
and substitute them into the Bessel Equation (1):
n =0
n =0
n =0
n =0
(n + )(n + 1)cn x n+ + (n + )cn x n+ + cn x n+ +2 2 cn x n+
=0
Divide the equation by x and collect the terms
n =0
n =0
n(n + 2 )cn x n + cn x n+2 = 0

Rename indices:
in the first sum m = n ; in the second sum n = m + 2
m=0
m=2
m(m + 2 )cm x m + cm2 x m = 0
Combine both series:
0 c0 + (1 + 2 )c1 x + [m(m + 2 )cm + cm2 ]x m = 0
(16)
m=2
Applying the Identity Theorem 2.6 to the term with summation, we obtain
a recurrence relationship:
cm =
cm 2
m(m + 2 )
for m 2
Using this relationship and also the first two terms of the equation, we get:
m=0
0 c0 = 0
m=1
(1 + 2 )c1 = 0
m=2
c2 =
m=3
m=4
c0
c0
=
2(2 + 2 ) 2 2(1 + )
c1
c3 =
=0
3(3 + 2 )
c0
c2
c4 =
=
4(4 + 2 ) 2 2 2 2 2(1 + )(2 + )
c0 = arbitrary
(by assumption, > 0 )

c0
c2 =
2 2(1 + )
c1 = 0
c3 = 0
c4 =
c0
2 4 2(1 + )(2 + )
m=5
m=6
c3
=0
5(5 + 2 )
c4
c4
=
c6 =
6 (6 + 2 ) 2 2 3(3 + )
c5 = 0
c5 =
c6 =
c0
2 3! (1 + )(2 + )(3 + )
4
All coefficients with odd indices are equal to zero. Recognizing the
pattern, we can determine the coefficients with even indices:
( 1)k c0
c2 k = 2 k
k = 0 ,1,2 ,...
2 k ! (1 + )(2 + ) (k + )
This expression may be written in more elegant form if the gamma
function is used. Multiply and divide the expression by (1 + ) :
( 1)k c0
(1 + )
2 2 k k ! (1 + )(1 + )(2 + ) (k + )
Repeatedly using the property (a) of the gamma function, we squeeze the
product in the denominator:
(1 + )(1 + )(2 + ) (k + ) = (2 + )(2 + )(3 + ) (k + ) = (k + + 1)
Then the expression for the coefficients becomes:
( 1)k c ( + 1)
c2 k = 2 k 0
2 k ! ( + k + 1)
1
Choose the value for the arbitrary coefficient c0 =
, then
2 ( + 1)
c2 k =
c2 k =
( 1)k
2 2 k + k ! ( + k + 1)
Then the solution becomes
( 1)
2 k +
x 2 k +
2
=
k
!
+ k + 1)
(
)
+
+
2
k
!
k
1
k =0
k =0
This series solution converges absolutely for all x 0 because there are
no other singular points. The function represented by this power-series
solution is called to be a Bessel function of the 1st kind of order and it
is denoted by
y1 ( x ) =
Bessel function of the 1st kind
( 1)k x
2 k +
( 1)k x
2 k +
2
(
k
!
+ k + 1)
k =0
J ( x ) =
(17)
This formula is valid for any real 0 (including integers = n and

2n + 1
half of the odd integers =
).
2
If is integer (let = n ), then the gamma function is replaced by the
factorial
(n + k + 1) = (n + k )!
and the solution simplifies to:
J n (x ) =
k =0
( 1)k x
2 k +n
2
k ! (n + k )!
(18)
This is a Bessel function of the 1st kind of integer order (including zero).
5. Second Solution
Case 1) r2 = int eger

2n + 1
)
2
Because x appears squared in the equation, the second solution may be
obtained from the first by a simple replacement of by :
(including halfs of the odd integers =
( 1)k x
2 k
2
k
!
+ k + 1)
k =0
J (x ) =
(19)
Functions J (x ) and J (x ) are linearly independent. It can be shown

that the Wronskian of Bessel functions (17) and (19) is:
W [J (x ), J (x )] =
J (x ) J (x )
2 sin
=
J (x ) J (x )
x
(20)
If n is not integer, then the Wronskian is not zero and the Bessel
functions J (x ) and J (x ) are linearly independent.
Then the general solution of the Bessel Equation may be written as
y (x ) = c1 J (x ) + c2 J (x )
Case 2) = n
(21)
When is integer, the Wronskian (20) is equal to zero
for any x > 0 , therefore, Bessel functions of integer orders J n (x ) and

J n (x ) are linearly dependent.
We can show that in this case functions J n (x ) and J n (x ) are just
multiples of each other. Indeed, write a Bessel function of negative
integer order replacing by n in equation (19):
2 k n
( 1) x
2
(22)
J n (x ) =
k =0 k! ( n + k + 1)
and change the index of summation k by s to make substitution in the
exponentiation
2k n = 2 s + n
then k = s + n , and equation (22) becomes
k
J n ( x ) =
( 1)s+n x
2 s+n
(s + n )!(s+ 1)
(23)
s= n
Consider a factor in the denominator (s + 1) : when s + 1 0 (nonpositive integer), the gamma function is unbounded, therefore, the first n
terms from s = n to s = 1 in the summation (23) are equal to zero,
then taking that into account for integers, (s + 1) = s! , we obtain
J n (x ) =
( 1)s+n x
2
(s + n )!s!
s =0
or
2 s+n
J n (x ) = ( 1) J n (x )
n
== ( 1)
( 1)s x
2 s+n
(s + n )!s!
s =0
(24)
So, function J n (x ) is the function J n (x ) up to the sign.

We need to find the second linearly independent solution.
According to the Frobenius Theorem 2.11, it can be found in the form:
J (x ) = d k x k + cJ (x ) ln x
k =0
Bessel function of the 2nd kind
or we can use the reduction formula (Part I, 2.2.9 3, p.131) to find the
second solution:
dx
J (x ) = J (x ) 2
xJ (x )
where long division and the Cauchy product should be used (which is
tedious but manageable).
But, instead, traditionally, the second independent solution is introduced
by the definition of the Bessel function of the 2nd kind of order :
Y (x ) =
cosJ (x ) J (x )
sin
for n not integer
(25)
and for integers, as the limit

Yn (x ) = lim Y (x ) =
n
cosJ (x ) J (x )
sin
(26)
which appear to exist for all n = 0,1,2,... (or n Z ).

Can be derived:
2 x
Yn (x ) = ln + J n (x ) .......
(27)
2
1
1 1
= lim 1 + + + + ln m = 0.5772... Eulers constant

m
m
2 3
Functions Yn (x ) have a logarithmic singularity at zero, while functions

J n (x ) are finite at zero, that leads to their linear independence. It can be
shown that the Wronskian for these functions is given by
J (x ) Y (x ) 2
=
(28)
W [J (x ), Y (x )] =
J (x ) Y (x ) x
General solution of Bessel equation
Functions J (x ) and Y (x ) are linearly independent for all (including

integers), and can be used for construction of the general solution of the
Bessel equation:
y (x ) = c1 J (x ) + c2Y (x )
(29)
When the order of the Bessel equation is not integer, the complete
solution may be also given only in the terms of Bessel functions of the
first kind:
y (x ) = c1 J (x ) + c2 J (x )
(30)
The second solution was derived mostly for integer roots, so, we
emphasize it by the following statement: the complete solution of the
Bessel equation of integer order n is given by:
y (x ) = c1 J n (x ) + c2Yn (x )
6. Bessel functions of half orders
(31)
It is happens that functions of orders = n

of elementary functions. Show it for =
1
can be expressed in terms
2
1
. Consider BE (14).
2
Use substitution y =
= x 2u
3
y =
du
1 2
x u+x 2
2
dx
5
du
du
3 2
x ux 2
+x 2
4
dx
dx
then the equation becomes
1
2
4
u + 1 +
u = 0
x2
y =
1
, this equation reduces to a linear ODE with constant
2
coefficients
u + u = 0
the general solution of which is given by
For =
y = c1 cos x + c 2 sin x
Apply the back substitution u = y x , then the solution becomes

y = c1
cos x
x
sin x
+ c2
If we choose for constants c1 = c 2 =
to be consistent with definition
(17), we obtain that Bessel functions of half orders are given by:
J 1 (x ) =
2
1
2
(x ) =
2 cos x
(32)
2 sin x
(33)
It can be verified that Bessel functions of =
3
are:
2
sin x
cos x
x
J 3 (x ) =
x
2
2
(34)
cos x
+ sin x
x
J 3 (x ) =
(35)
x
2
Other Bessel functions of half of odd integer orders also can be expressed
in terms of elementary functions. These functions are used for
construction of spherical Bessel functions
J 1 (x )
n
n+ 2
1 d sin x
= xn
(36)
j n (x ) =
x
2
x dx
x
2
1
2
(x )
1 d cos x
= x n
(37)
x
2
x dx
x
which are solutions of the Bessel equation
x 2 y + 2 xy + x 2 n(n + 1) y = 0 n = 0,1,2,...
(38)
This equation appears as one of the ODE in the separation of variables of
the Laplacian in spherical coordinates [Abramowitz and Stegun, p.437]
y n (x ) =
n+
7. Bessel functions of the 3rd kind
A linear combination in the general solution (29) assumes that coefficients

c1 and c2 are real numbers. We also considered the variable x to be a
real number too. But the obtained equations and functions are valid also
for complex numbers. Two particular combinations of Bessel functions
J v (z ) and Y (z ) with complex coefficients lead to the introduction of the
complex version of Bessel functions, which also are the solutions of the
Bessel equation but in the field of complex numbers z Z :
We define two Bessel functions of the 3rd kind of order ( they are
also called Hankel functions) as
H(1) (z ) = J (z ) + iY (z )
(39)
H(2 ) (z ) = J (z ) iY (z )
(40)
or we can express them in terms of the Bessel function J (z ) only if

function Y (z ) is replaced in this definition by a complex version of
equation (25):
H(1) (z ) =
J (z ) e i J (z )
i sin
(41)
H(1) (z ) =
J (z ) ei J (z )
i sin
(42)
Definitions (41) and (42) are for v integer . For = n , n = 0,1,2,... , we

take the limits:
J (x ) e i J (x )
(43)
H n(1) (z ) = lim
n
i sin
J (x ) ei J (x )
(44)
H n(2 ) (z ) = lim
n
i sin
Because Hankel functions H(1) (z ) and H(2 ) (z ) are linear combinations of
Bessel functions of the 1st and the 2nd type, they have the similar
properties.
The Wronskian of Hankel functions H(1) (z ) and H(2 ) (z ) is
4i
W H(1) (z ), H(2 ) (z ) =
z
therefore functions H(1) (z ) and H(2 ) (z ) form a fundamental set for the
Bessel equation; and the general solution of the Bessel equation can be
written as
y (x ) = c1 H(1) (z ) + c2 H(2 ) (z )
for any order of the Bessel equation (including integers).
8. Properties of Bessel functions
a) Functions J (x ) and Y (x ) are both oscillatory; they have infinitely

many roots for x > 0 .
x2 x4
x6
x8
+
+
+
4 64 2304 147456
x x3
x5
x7
+
+
J1 (x ) =
2 16 384 18432
x2 x4
x6
x8
+
J 2 (x ) =
8 96 3072 184320
J 0 (x ) = 1
Identities
Differential identities
2
J (x ) J 1 (x )
x
2
J 1 (x ) =
J (x ) J +1 (x )
x
J +1 (x ) =
Addition theorem
(46)
d
x J (x ) = x J 1 (x )
dx
d
x J (x ) = x J +1 (x )
dx
J 0 (x ) = J 1 (x )
xJ (x ) + J (x ) = xJ 1 (x )
J (x ) = J 1 (x )
xJ (x ) J (x ) = xJ +1 (x )
J (x ) = J +1 (x ) +
Integral identities
(45)
(47)
(48)
+1
+1
x J (x )dx = x J +1 (x ) + c
+1
+1
x J (x )dx = x J 1 (x ) + c
J n (x + y ) =
J (x )
J (x )
(49)
(50)
(51)
(52)
J k (x )J n k ( y )
(53)
k =
The same properties also hold for Y (x ) .

b) Functions I (x ) and K (x ) are not oscillatory
x2 x4
x6
x8
+
+
+
+
4 64 2304 147456
x x3
x5
x7
+
+
+
I1 ( x ) = +
2 16 384 18432
x2 x4
x6
x8
+
+
+
+
I 2 (x ) =
8 96 3072 184320
I 0 (x ) = 1 +
9. Generating functions
1 x
t
t 2
J n (x )t n
(55)
n =
cos x = J 0 (x ) 2 J 2 (x ) + 2 J 4 (x ) 2 J 6 (x ) +
sin x = 2 J 1 (x ) 2 J 3 (x ) + 2 J 5 (x ) +
(What can be obtained for x=1 ?)

Expansions of xn
1 = J x (x ) + 2 J 2 (x ) + 2 J 4 (x ) +
+ 2 J 2 k (x ) +
1
x = J 1 (x ) + 3J 3 (x ) + 5 J 5 (x ) +
2
+ (2k + 1)J 2 k +1 (x ) +
1
x = J 2 (x ) + 4 J 4 (x ) + 9 J 6 (x ) +
8
+ k 2 J 2 k (x ) +
10. Modified Bessel equation
The modified Bessel equation is given by
x 2 y + xy x 2 + 2 y = 0
(56)
which can be written in the form of Bessel equation (14) with the second
parameter = i :
x 2 y + xy + i 2 x 2 2 y = 0
(57)
which has a general solution given by equation (29) with x replaced by ix:
y (x ) = c1 J (ix ) + c 2 Y (ix )
(58)
Equation (58) provides a solution of the modified Bessel equation (91) in
complex form. But it is desirable to have a real form of solution.
Consider
J (ix ) =
k =0
( 1)k ix
2 k +
2 k +
2
2
=i
k! ( + k + 1)
k = 0 k! ( + k + 1)
(59)
because:
( 1)k (i )2 k + = ( 1)k (i )2 k i = ( 1)k (i 2 )k i = ( 1)k ( 1)k i = ( 1)2 k i
= i .
Then define a function which is called the modified Bessel function of
the 1st kind of order :
2 k +
I (x ) = i J (ix ) =
k = 0 k! ( + k + 1)
(60)
which is a real function and which is a solution of the modified Bessel

equation (56). Notation I for this function reflects the method of its
definition, and it means the function of imaginary argument. For
negative values of parameter , define a second solution of the
modified Bessel equation as
2 k
I (x ) = i J (ix ) =
k = 0 k! ( + k + 1)
(61)
The Wronskian of functions I (x ) and I (x ) can be calculated as

2 sin
0 for v integer
x
therefore, functions I (x ) and I (x ) are linearly independent and form a
fundamental set, then the general solution of the modified Bessel equation
of non-integer order is given by
W [I (x ), I (x )] =
y (x ) = c1 I (x ) + c 2 I (x )
v integer
(62)
In the case of integer orders, function I n (x ) is the same as function

I n (x ) . Indeed, when is changed for n in equation (61)
I n (x ) = i n J n (ix ) = i n ( 1) J n (ix ) = i n ( 1) i n i n J n (ix )

n
( )
= i ( 1) I n (x ) = ( 1) ( 1) I n (x ) = I n (x )
For integer orders = n , n = 0,1,2,3, , the second solution of the
modified Bessel equation is defined with the help of the modified Bessel
function of the 2nd kind of order :
2 n
K (x ) =
I (x ) I (x )
2
sin
(63)
as the limit of
K n (x ) = lim K (x )
(64)
11. Equations solvable in terms of Bessel functions
Consider some generalizations of the Bessel equation which also can be

solved in terms of the Bessel function.
a) Consider a generalized Bessel differential equation of the form
(2m 1) m 2 p 2 2
1 2m
y +
2 y + p 2 a 2 x 2 p 2 + 2 +
+
y = 0
x
x2
x
if y = y (x ) is any solution of the Bessel equation, then the function
( )
y = x m e x y ax p
is a solution of the generalized equation. For instance, for any real 0
(including integers), the general solution can be written as
y = x m e x c1 J ax p + c 2 Y ax p
or for non-integer orders, a general solution can be written as
y = x m e x c1 J ax p + c 2 J ax p
Proof of this statement can be made by the appropriate change of variables
and by reduction of the differential equation to the Bessel equation.
Example 1
( )
( )]
( )
( )]
Check that the modified Bessel equation

x 2 y + xy x 2 + 2 y = 0
is a particular case of the generalized equation. Rewrite
it in the form of the generalized equation:
2
1
y + y + 1 2 y = 0
x
x
0 1 2
1 2 0
y +
2 0 y + (1)( 1)x 21 2 +
y = 0
x2
x
from which we can identify

m = 0 , = 0 , p = 1 , and a 2 = 1
and, therefore, solutions of the modified Bessel equation
should include functions
J (ix ) , J (ix ) , and Y (ix )
what we know from Section 10.
Airy equation
Example 2
Consider the Airy equation

y xy = 0
which is the simplest case of the linear 2nd order ODE
with variable coefficients. This equation has
applications in dynamics (oscillation of an aging
spring), quantum mechanics and optics.
Rewrite the Airy equation in the form of the generalized

equation
1
1 9 2

1 2 2
9 4 2 3 2
2 0 y + x 2 + 0 + 0 + 4 4 y = 0
y +
x
x
4 9
from which we can identify

4
1
1
3
m = , = 0 , p = , a 2 = , and 2 =
2
2
9
9
from the last equation we can determine the order of the
equation
1
=
3
Then solutions of the Airy equation can be written as
1
2 3
y1 (x ) = x 2 J 1 ix 2
3
1
2 3
y 2 (x ) = x 2 J 1 ix 2
3
3
If we rewrite Bessel functions of the 1st kind of complex

arguments in terms of modified Bessel functions using
equation (95), then the solutions become
2 3
y 1 (x ) = x I 1 x 2
3
2 3
y 2 (x ) = x I 1 x 2
3
3
These two linearly independent solutions (note, that

order of modified Bessel functions is not integer) may
be used for construction of the traditional form of
solutions
3
2 3
x 2 2
Ai(x ) =
x I 1 x 2
I 1
3 3 3
33
3
3
2
x 2 2
Bi(x ) =
x + I 1 x 2
I 1
3 3 3
33
which are called Airy functions. The next plot shows
the graph of Airy functions
Bi ( x )
Ai ( x )
b) This form of equation is a particular case of the previous equation, but

it is more convenient for applications in simpler cases:
y + a 2 x p 2 y = 0
solutions of this equation have the forms
2a p
x2
y (x ) = x J 1
p
p
2a p
y (x ) = xY 1
x2
p
p
Solutions of the Airy equation can be obtained in this case much faster.
c) Equation
y + a 2 e 2 x p 2 y = 0
has solutions
y (x ) = J p ae x
( )
(ae )
(ae )
y (x ) = J p
y (x ) = Y p
12. Orthogonality of Bessel functions
We know from Sturm-Liouville theory that solutions of the self-adjoint

differential equation satisfying homogeneous boundary conditions form a
complete set of functions orthogonal with some weight function (SturmLiouville theorem). Consider application of this theory to solutions of
BVP for BE.
Self-adjoint form of BE
The Bessel equation of order with parameter

x 2 y + xy + 2 x 2 2 y = 0
can be reduced to a self-adjoint form with the help of a multiplying factor
(x ) =
a1
1 a0 dx
1 2 dx
1
1
e
= 2 e x = 2 e ln x =
a0
x
x
x
After multiplication of BE by
+ 2 x y = 0
x
[xy ] +
1
it can be reduced to self-adjoint form
x
(identify p(x ) = x )
Then, the Sturm-Liouville Problem in the interval x [ L1 ,L2 ] produces

infinitely many values of the parameter n (eigenvalues) for which there
exist non-trivial solutions yn (x ) (eigenfunctions).
According to the Sturm-Liouville theorem, the obtained eigenfunctions
are orthogonal with the weight function p(x ) = x :
L2
xyn ( x )ym ( x ) dx = 0
for n m
L1
Orthogonal sets for circular domain
Consider BE in the finite circle 0 x L . The general solution is given

by
y (x ) = c1 J (x ) + c2Y (x )
0xL
The physical sense of solution of classical PDE requires a finite value of

solution in all points of [0 , L] . Bessel functions of the second kind Y (x )
are unbounded at x = 0 , therefore, to satisfy this condition we have to put
an arbitrary constant c 2 equal to zero. Then solution of BE becomes
y (x ) = c 1 J (x )
boundary conditions:
Consider the homogeneous boundary conditions at x = L :

I
Dirichlet y (x ) x = L = 0
II
Neumann
III Robin
y (x ) x = L = 0
[y (x ) + Hy(x )]x = L
=0
We are looking for the values of the parameter which provide nontrivial solutions y (x ) = c 1 J (x ) satisfying boundary conditions I-III.
These values can be found by substitution of the solution into the
boundary conditions I-III as the positive roots of the following equations:
Equations for eigenvalues
J (L ) = 0
II
J +1 (L ) +
III
J (L ) = 0
L
( 0 = 0 is also an eigenvalue when = 0 )
d
J (x ) = 0
J (x ) = 0
dx
J +1 (x ) +
J (x ) = 0
x
J +1 (L ) + H + J (L ) = 0
L
Proof:
y (x ) + Hy (x )
d
=
J (x ) + HJ (x )
( = J (x ) + HJ (x ) , Oz)
dx
J (x ) + HJ (x )
= J +1 (x ) +
x
= J +1 (x ) + H + J (x )
x
In the particular case, when the Bessel function is of zero order, = 0 ,

equations for eigenvalues are:
I
II
III
Orthogonality
J 0 (L ) = 0
J 1 (L ) = 0
(and 0 = 0 is also eigenvalue)
J 1 (L ) + HJ 0 (L ) = 0
Obtained equations generate infinitely many eigenvalues n , n = 1,2 ,3 ,...

For which the corresponding eigenfunctions are:
yn ( x ) = { J ( n x )}
The corresponding set of solutions {J ( n x )} is orthogonal with respect

to the weight function p(x ) = x :
L
xJ ( n x )J ( m x )dx =
0
0
2
N
nm
n=m
where the squared norm of eigenfunctions is determined as

(Ozisik N. Heat Transfer, p.133; McLachlan Bessel Functions for
Engineers, p.110):
Norm of eigenfunctions
2
,n
= xJ ( n x )dx
2
x2
=
2
=
L2
2
2
2
J ( n x ) + 1
2 x 2
2
2
J ( n L ) + 1
2 L2
2
J ( n x )
2
J ( n L )
or integrating with Maple:

L
x2
= J2 ( n x ) J 1 ( n x ) J +1 ( n x )
2
0
0
The derivative can be expressed as (use chain rule and identity in sec. 6)
d
J ( n x )
=
J ( n x )
dx
J ( n x )
= n J +1 ( n x ) +
n x
N2,n
= xJ 2 ( n x )dx
= n J +1 (n x ) +
J (n x )
x
or if we use the other identity for lower order then
d
J ( n x )
=
J ( n x )
dx
J ( n x )
= n J 1 ( n x )
n x
= n J 1 ( n x )
J ( n x )
Then taking into account that eigenvalues n satisfy equations I-III (that
simplifies expressions), the squared norm for specific boundary conditions
is given:
Bessel-Fourier Series
N2,n
II
N2,n
III
N2,n
L2 2
J +1 (n L ) or
2
2
L2
1 2 2 J2 (n L )
=
2
n L
L2
2
H2
2
2 + 1 2 2
n L
n
L2 2
J 1 ( n L )
2
2
L2
N 0 ,0 =
2
J (n L )
The obtained orthogonal systems can be used for constructing the function
expansion in a generalized Fourier series
f ( x ) = an J ( n x )
n =1
where coefficients c n are determined from the equation

L
an =
xJ ( n x ) f ( x ) dx
0
2
xJ ( n x ) dx
0
xJ ( n x ) f ( x ) dx
0
N2,n
Example 3 (I Dirichlet boundary condition)
y ( L) = 0
Consider an orthogonal set obtained as a solution of a Dirichlet problem
with Bessel functions of zero order yn( ) ( x ) = J ( n x ) where
eigenvalues n are positive roots of equation J ( L ) = 0 .
The squared norm of eigenfunctions can be calculated as

L2 2
N2,n
=
J +1 ( n L )
2
Then expansion of function f (x ) in Fourier-Bessel series has the form
L
f ( x ) = an J 0 ( n x ) ,
n =1
where an =
2
L2
xJ ( n x ) f ( x ) dx
0
J2+1 ( n L )
(it is also known as the Hankel series (1869)).

Consider now expansion of the function f (x ) = 1 H (x 1) , x [0 ,3] in
the Hankel series of order = 0 . Coefficients are
3
an =
xJ 0 ( n x ) dx
0
2
0 ,n
J 1 ( n ) 1
2 J 1 ( n )
=
n N02,n L2 n J 12 ( n L )
and the expansion becomes

f ( x) =
2 J 1 ( n ) J 0 ( n x )
9 n =1 n J 12 ( 3n )
This example can be illustrated with a Maple presentation (SF-1.mws)
Example 4 (II Neumann boundary condition)
Consider an orthogonal set obtained as a solution of the Neumann

problem with Bessel functions y n( ) (x ) = J ( n x ) where eigenvalues
n are positive roots of the equation
y ( L ) = 0
J +1 (L ) +
J (L ) = 0 ( 0 = 0 is also an eigenvalue for = 0 )

L
L2
L2
2
1 2 2 J2 (n L ) and N 02,0 =
=
N2,n
2
2
n L
Fourier-Bessel series:
=0
0
f ( x ) = a0 + an yn( ) ( x ) ,
where
n =1
an =
a0 =
xJ 0 ( n x ) f ( x ) dx
0
N2,n
2 L
xf ( x ) dx
L2 0
L
>0
f ( x ) = an yn( ) ( x ) ,
n =1
where
an = xJ ( n x ) f ( x ) dx
0
N2,n
Consider now expansion of the function f (x ) = 1 H (x 1) , x [0 ,3] ,

=0:
3
an =
xdx
0
2
0 ,0
1
9
an =
xJ 0 ( n x ) dx
0
2
0 ,n
L2 J 1 ( n )
2 J 1 ( n )
=
2
2 n J 0 ( n L ) 9 n J 02 ( 3n )
Maple solutions:
=0
=1
SF-1-2-0.mws
SF-1-2-1.mws
Example 5 (III Robin boundary condition)
y ( L ) + Hy ( L ) = 0
Consider an orthogonal set obtained as a solution of the Robin problem
with Bessel functions of zero order yn( ) ( x ) = J ( n x ) where eigenvalues
n are positive roots of equation

0
J +1 ( L ) + H + J ( L ) = 0
L

2 2
L2 H 2
N2,n
=
2 + 1 2 2 J (n L )
2 n
n L
L
f (x )
= an yn( ) ( x )
n =1
where an =
xJ 0 ( n x ) f ( x ) dx
0
N2,n
Consider expansion of function f (x ) = 1 H (x 1) , x [0 ,3] , H = 2

Maple solution (for = 1 )
SF-1-3.mws
SF-1.mws Example 3 Fourier-Bessel series

> nu:=0;
I Dirichlet boundary condition = order of Bessel Functions
:= 0
> L:=3;
L := 3
> f(x):=1-Heaviside(x-1);
f( x ) := 1 Heaviside ( x 1 )
Characteristic equation:
> w(x):=BesselJ(nu,x*L);
w ( x ) := BesselJ ( 0 , 3 x )
> plot(w(x),x=0..10);
Eigenvalues:
> lambda:=array(1..200);
:= array ( 1 .. 200 , [ ] )
> n:=1: for m from 0 to 50 do

y:=fsolve(w(x)=0,x=m/2..(m+1)/2): if type(y,float) then
lambda[n]:=y: n:=n+1 fi od:
> for i to 2 do lambda[i] od;
0.8016085192
1.840026037
> N:=n-1;n:='n':i:='i':m:='m':y:='y':x:='x':
N := 24
Eigenfunctions:
y[n]:=BesselJ(nu,lambda[n]*x);
yn := BesselJ ( 0 , n x )
Squared Norm:
>NY[n]:=int(x*y[n]^2,x=0..L):
NY[n]:=subs(BesselJ(nu,L*lambda[n])=0,NY[n]):
Fourier-Bessel coefficients:
a[n]:=int(x*y[n]*f(x),x=0..L)/NY[n];
a n :=
BesselJ ( 1 , n )
2
9 BesselJ ( 1 , 3 ) 2
n
n
> u(x):=Sum(a[n]*y[n],n=1..N);
24
u( x ) :=
n=1
2 BesselJ ( 1 , n ) BesselJ ( 0 , n x )
n BesselJ ( 1 , 3 n )
> u(x):=sum(a[n]*y[n],n=1..N):
> plot({f(x),u(x)},x=0..L,discont=true,color=black);
Orthogonal sets for annular domain

L1 x L2
L1
L2
Consider BE in the finite circle

x 2 y + xy + 2 x 2 2 y = 0 , x (L1 , L2 )
with homogeneous boundary conditions:
dy
h
H1 = 1
=0
k1 dx + h1 y
k1
x= L1
h
dy
=0
H2 = 2
k 2 dx + h2 y
k2
x = L2
The general solution is given by
y (x ) = c1 J (x ) + c2Y (x )
A BVP for BE in the finite domain according to the Sturm-Liouville
theorem generates an infinite set of eigenvalues n and corresponding
eigenfunctions y n (x ) orthogonal with the weight function p(x ) = x . A

particular form of the orthogonal set depends on the type of boundary
conditions.
Consider a case when both boundary conditions are of Dirichlet type:
Example 6
(Dirichlet-Dirichlet boundary conditions)
Boundary conditions:
y x= L = 0
1
y x= L = 0
2
Apply boundary conditions to the general solution of BE:

c1 J (L1 ) + c2Y (L1 ) = 0
c1 J (L2 ) + c2Y (L2 ) = 0
This is a homogeneous system of two linear algebraic equations for c1
and c2 . Rewrite it in the matrix form
J (L1 ) Y (L1 ) c1 0
J (L ) Y (L ) c =
2
2 2
0

We are looking for non-trivial solution of BVP, i.e. both coefficients in
general solution cannot be zero
c1 0
c
2 0
A homogeneous linear system has a non-trivial solution only if the

determinant of the system matrix is equal to zero:
Equation for eigenvalues n :
J (L1 ) Y (L1 )
det
= J (L1 )Y (L2 ) J (L2 )Y (L1 ) = 0
J (L2 ) Y (L2 )
The roots of this equation yield the eigenvalues n for which BVP has
non-trivial solutions y n (x ) (eigenfunctions). Oscillatory property of
Bessel functions provides an infinite set of eigenvalues n and
corresponding eigenfunctions are
yn (x ) = c1,n J (n x ) + c2 ,nY (n x )
Determine now the coefficients c1,n and c2 ,n from a system where
eigenvalues are substituted
J (n L1 ) Y (n L1 ) c1,n 0
J ( L ) Y ( L ) c =
n 2 2 ,n
0
n 2
Because a linear system has a singular matrix, solutions for c1,n and c2 ,n
are linearly dependent and can be determined just from one equation, let it
be the second one
c1,n J (n L2 ) + c2 ,nY (n L2 ) = 0
one of the unknowns in this equation is a free parameter, choose
1
1
, then c2 ,n =
c1,n =
(
)
(
J n L2
Y n L2 )
Then eigenfunctions have the form:
y n (x ) =
Eigenfunctions
J (n x )
Y ( x )
n
J (n L2 ) Y (n L2 )
The norm of eigenfunctions is given by:

N2,n
L2
= xyn2 (x )dx
L1
J ( x ) Y ( x )
= x n n dx
Y (n L2 )
L1 J (n L2 )
2
L2
L2
L2
xJ (n x )dx + 2
xY (n x )dx
J (n L2 ) L
Y (n L2 ) L
2
2
1
xJ (n x )Y (n x )dx
J (n L2 )Y (n L2 ) L1
= ... express in terms of J +1 ,
Summary:
For an annular domain with boundary conditions:

y x= L = 0
1
y x= L = 0
2
Eigenvalues n are positive roots of the characteristic

equation
J (L1 )Y (L2 ) J (L2 )Y (L1 ) = 0
The eigenfunctions are
J ( x )
Y ( x )
y n (x ) = n n
J (n L2 ) Y (n L2 )
f ( x ) = an y n ( x )
n =1
where
an =
xyn (x ) f (x )dx xJ (n x ) f (x )dx

0
xy (x )dx
2
n
N2,n
Maple examples:
=0
=1
SF-AD-1-0.mws
SF-AD-1-1.mws
L1 = 2 , L2 = 5 f (x ) = 1 H (x 3 )
Consider a case when both boundary conditions are of Robin type:

Example 7 (Robin-Robin boundary conditions)
Consider BE in the annular domain

x 2 y + xy + 2 x 2 2 y = 0 , x (L1 , L2 )
with homogeneous boundary conditions:
dy
h
=0
H1 = 1
k1 dx + h1 y
k
x= L1
1
x = L1
x = L2
x = L1
x = L2
L1
L2
SF-5.mws
h
dy
=0
H2 = 2
k 2 dx + h2 y
k
x = L2
2
The general solution is given by
y (x ) = c1 J (x ) + c2Y (x )
The derivative of the general solution (use chain rule and differential
identities)
d
y ( x ) = c1 J + 1 ( x ) +
J ( x ) + c2 Y + 1 ( x ) +
Y ( x )
dx
x
x
Substitute into boundary conditions:
c1 J +1 (L1 ) +
J (L1 ) c2 Y +1 (L1 ) +
Y (L1 ) + c1 H 1 J (L1 ) + c2 H 1Y (L1 ) = 0
L1
L1
c1 J +1 (L2 ) +
J (L2 ) + c2 Y +1 (L2 ) +
Y (L2 ) + c1 H 2 J (L2 ) + c2 H 2Y (L2 ) = 0
L2
L2
Collect terms

c1 J +1 (L1 ) + H 1 J (L1 ) + c2 Y +1 (L1 ) + H 1 Y (L1 ) = 0
L1
L1
J (L2 ) + c2 Y +1 (L2 ) + H 2
c1 J +1 (L2 ) + H 2 +
L
L
2
2
Denote:

a11 = J +1 (L1 ) + H 1 J (L1 )
L1
Y (L2 ) = 0

a12 = Y +1 (L1 ) + H 1 Y (L1 )
L1

a21 = J +1 (L2 ) + H 2 + J (L2 )
L2

Y (L2 )
a22 = Y +1 (L2 ) + H 2
L2
Then a system for coefficients has the following matrix form:

a11 a12 c1 0
a
=
21 a22 c2 0
A necessary condition for a system to have a non-trivial solution is

a12
a
det 11
=0
a
a
22
21
it yields a characteristic equation for values of the parameter for which
the BVP has a non-trivial solution:
Equation for eigenvalues n :

J +1 (L1 ) + H 1 J (L1 )
L1

Y + 1 (L2 ) + H 2 + Y (L2 )
L2

J (L 2 ) = 0
Y +1 (L1 ) + H 1 Y (L1 ) J +1 (L2 ) + H 2
L1
L2
The positive roots of this equation provide an infinite set of eigenvalues

n . Then for the determined eigenvalues n , coefficients c1,n and c 2 , n
can be found from one of the equations of the system (choose the second
one):
a21c1 + a22 c2 = 0
One of the coefficients can be taken as a free parameter, choose
1
1
, then c2 =
c1 =
a21
a22
With determined coefficients,
(eigenfunctions) have the form:
Eigenfunctions:
y n (x )
solutions
of
the
BVP
J (n x ) Y (n x )
a21,n
a22 ,n
J (n x )
n J +1 (n L2 ) + H 2 +
L2
J (n L2 )
nY +1 (n L2 ) + H 2
L2
Y (n L2 )
Y (n x )
The norm of the eigenfunctions is determined by the integral

L2
= xyn2 (x )dx
N2,n
L1
f ( x ) = an y n ( x )
n =1
L2
L2
L1
L1
xy n (x ) f (x )dx
where a n =
L2
xy (x )dx
xy n (x ) f (x )dx
2
n
N2,n
L1
Maple example
L1 = 2 , L2 = 5
H1 = 2 , H 2 = 3
f (x ) = 1 H (x 3)
=0
=1
SF-AD-9-0.mws
SF-AD-9-1.mws
y n (x )
5.7 Legendre Functions

Separation of variables of the Laplacian in a spherical coordinate system
yields a group of ODE one of which has the form

m2
1 x 2 y + n(n + 1)
y = 0
1 x2
where m and n are separation constants.

This equation is called Legendres associated differential equation.
Solution of this equation includes Legendres associated functions of
degree n and of order m of the 1st and the 2nd kind Pnm (x ) and Q nm (x ) .
When m = 0 (in a case when the Laplacian does not depend on the
variable ), equation is called the Legendres differential equation
1 x 2 y + n(n + 1) y = 0
Solution of this equation include Legendres functions of degree n of the
1st and the 2nd kind Pn (x ) and Q n (x ) .
1. Legendre Equation
[(
) ]
[(
) ]
Consider the Legendre differential equation rewritten in standard form

1 x 2 y 2 xy + n(n + 1) y = 0
xR
This equation has two singular points x = 1 , all other points are ordinary
points. We will apply a power-series solution method around the ordinary
point x = 0 (the interval of convergence for this solution is (1,1) ).
Assume that the solution is represented by a power series
2. Solution of Legendre Equation
y = ak x k
k =0
then derivatives of the solution are
y = ka k x k 1
k =1
y = k (k 1)a k x k 2
k =2
Substitute them into equation
( 1 x ) k ( k 1) a x
2
k =2
k =2
k =2
k 2
2x kak x k 1 + n ( n + 1) ak x k = 0
k =1
k =0
k ( k 1) ak x k 2 k ( k 1) ak x k 2kak xk + n ( n + 1) ak x k = 0
(change k = k 2 in the first term)
k =0
k =0
k =2
k =1
k =0
( k + 1)( k + 2 ) ak + 2 x k k ( k 1) ak x k 2kak x k + n ( n + 1) ak x k = 0
2 1 a 2 + 2 3 a 3 x 2 a 1 x + n(n + 1)a 0 + n(n + 1)a 1 x +
k =1
n ( n + 1) a0 + 1 2 a2 + 2 3 a3 + n ( n + 1) 2 a1 x +
{a k [ k (k 1) 2k + n(n + 1)] + (k + 1)(k + 2 )a k + 2 } = 0
k =2
{a k [n(n + 1) k (k + 1)] + (k + 1)(k + 2 )a k + 2 } = 0
k =2
Using the comparison theorem, determine the relation for coefficients:

n(n + 1)
a2 =
a0
1 2
(n 1)(n + 2 ) a
2 n(n + 1)
a3 =
a1 =
1
23
23
k (k + 1) n(n + 1)
(n k )(n + k + 1)
ak =
a
k = 2 ,3,...
(k + 1) (k + 2 )
(k + 1) (k + 2 ) k
Coefficients a 0 and a 1 are arbitrary, consider them to be the parameters
for the general solution and collect the terms corresponding to these
coefficients, then the power series solution of the Legendre Equation
becomes
a k +2 =
y (x )
n ( n + 1) 2 n ( n 2 )( n + 1)( n + 3 ) 4 n ( n 2 )( n 4 )( n + 1)( n + 3 )( n + 5 ) 6
= a0 1
x +
x
x + ...
2!
4!
6!
( n 1)( n + 2 ) 3 ( n 1)( n 3 )( n + 2 )( n + 4 ) 5 ( n 1)( n 3 )( n 5 )( n + 2 )( n + 4 )( n + 6 ) 7

+ a1 x
x +
x
x + ...
3!
5!
7!
= a 0 L n ,1 (x ) + a 1 L n ,2 (x )
Choose a sequence of non-negative values of n = 0 ,1,2 ,... then

corresponding solutions are (note, that in the solution all terms except for
finite number alternatingly disappear: if n=2k is even then in the first
series all terms with multiple (n 2 k ) disappear, if n=2k+1 is odd then in
the second series all terms with multiple (n 2k 1) disappear, and they
become the finite polynomials)
n=0
L1,n (x )
L 2 ,n (x )
a0
a 1 L2 ,0 (x )
n=1
a 0 L1,1 (x )
n=2
a0 1 3 x 2
n=3
a 0 L1,3 (x )
a1 x
a 1 L 2 ,2 (x )
5
a1 x x 3
3
Choose
a 0 = ( 1) 2
n!
a 1 = ( 1)
n 1
2
(n + 1)!
2
n
n 1 n +1
2n !
2n
!
!
2

2 2
Then Legendre functions of the 1st kind for different values of parameter n
generate the following set of polynomials
P0 (x ) = 1
P1 (x ) = x
3 2 1
x
2
2
5 3 3
P3 (x ) = x x
3
2
35 4 15 2 3
x +
P4 (x ) =
x
8
4
8
which are called Legendre polynomials. Because Legendre polynomials

are solutions of the separated Laplace equation in spherical coordinates,
they are also called spherical harmonics (and the method of solution in
terms of Legendre functions is called correspondingly the Method of
Spherical Harmonics). Recall that this system of polynomials up to scalar
multiple was also obtained from orthogonalization of the linear
P2 (x ) =
independent set of monoms 1,x,x 2 ,x 3 ,... on the interval [ 1,1] .
Recurrence formula
(n + 1)Pn +1 (x ) = (2n + 1)xPn (x ) nPn 1 (x )
Rodrigues formula
Pn (x ) =
Orthogonality of Legendre polynomials
n
1 dn 2
x 1
n
2
2 n! dx
Legendre polynomials are orthogonal in the interval [ 1,1] with the

weight function p(x ) = 1
0
mn
1
Pm ( x ) Pn ( x )dx =
1
2
m=n
2n + 1
Fourier-Legendre series
Legendre polynomials can be used for expansion of the function f (x ) ,

x [1,1] in the Fourier-Legendre series:
f (x ) = c n Pn (x )
n =0
where expansion coefficients are

1
cn =
f (x )Pn (x )dx
Pn (x )dx
2
= n + f (x )Pn (x )dx
2 1
Example 8 (expansion in Fourier-Legendre series (spherical harmonics))

f (x ) = H (x ) x [1,1]
Maple Solution:
SF-8.mws
1 1
f (x ) = n + Pn (x )dx Pn (x )
2 0
n =0
Integral transform
The integral transform based on the Fourier-Legendre expansion
fn =
f (x )K n (x )dx
with inverse transform
f (x ) = f n K n (x )
n =0
where the kernel of the integral transform is defined as a normalized

Legendre function
1
K n (x ) = n + Pn (x )
2

SF-8.mws
Example 8 Fourier-Legendre Series
> restart;
> with(orthopoly);
[ G , H , L , P, T , U ]
> for n from 0 to 6 do P(n,x) od;

1
x
1 3 x2
+
2
2
5 3 3
x x
2
2
3 35 4 15 2
x
x
+
8 8
4
63 5 35 3 15
x
x +
x
8
4
8
5 231 6 315 4 105 2

x
x +
x
+
16
16
16
16
> f(x):=Heaviside(x);
f( x ) := Heaviside ( x )
Fourier-Legendre coefficients:
> c[n]:=(n+1/2)*int(f(x)*P(n,x),x=-1..1);
1
1
cn := n +
P ( n, x ) dx
0
Fourier-Legendre series:
> u(x):=sum(c[n]*P(n,x),n=0..10);
1 218295
315315 3 1702701 5 984555 7 1616615 9
u( x ) := +
x
x +
x
x +
x
2 65536
16384
32768
16384
65536
> plot({f(x),u(x)},x=-1..1);
> u(x):=sum(c[n]*P(n,x),n=0..100):
> plot({f(x),u(x)},x=-1..1);
The Best Approximation by Polynomials
Consider a vector space of square integrable functions L2 [ 1,1] . The span of all
polynomials of order n is a subspace of L2 [ 1,1] . Call it n .

Let f ( x ) n and let
n
f n ( x ) = cn Pn ( x )
k =1
be the nth partial sum of the Fourier-Legendre expansion of the function f ( x ) in
[ 1,1] .
Then f n ( x ) provides the best approximation of the function f ( x ) by the nth

order polynomials, i.e. function f n ( x ) is the closest to the function f ( x ) among
the functions in n in the sense that it minimizes the distance
f ( x ) pn ( x ) = ( f pn , f pn ) =
( f pn )
dx
Graphs of Legendre polynomials

Legendre-1.mws
Legendre polynomials
> restart;
> with(orthopoly):
> plot({P(0,x),P(1,x),P(2,x)},x=-1..1);
P0
P1
P2
> plot({P(3,x),P(4,x),P(5,x)},x=-1..1);
P3
P5
P4
> plot({P(6,x),P(7,x),P(8,x)},x=-1..1);
P6
P7
P8
5.8 Exercises:
1) Show
sin x (x n ) = 0
sin x (x n ) = ( 1) (x n )
n +1
(by multiplying both sides by an arbitrary differentiable function and

integrating)
2) Show
(3 x + 6 ) f (x )dx =
In general
f ( 2 )
3
f (x0 )
[g (x )] f (x )dx = g (x )
0
where g (x0 ) = 0
3) Solve the IVP and sketch the solution curves (use Maple and Laplace
transform):
y y = (t 2 )H (t 2 )
y (0 ) = 0
y (0 ) = 0
y + 2 y + y = 3 (t 2 )
y (0 ) = 1
y (0 ) = 1
4) Sign function is defined as
1 x >0
sgn(x ) =
1 x < 0
a) Express sgn(x ) in terms of Heaviside step function H (x )
b) Express Heaviside step function H (x ) in terms of sgn(x )
d
sgn(x )
dx
d) Sketch the graph of sgn(x 3)
c)
Calculate
5) Investigate convergence of the Bessel function J (x ) using the ratio

test (see supplemental materials)
6) Use term-by-term differentiation (why can we do it?) to show
d
J 0 (x ) = J 1 (x )
dx
7) Solve in terms of Bessel functions

x 2 y + (x 2 ) y = 0
8) Finish Examples 4 and 5 in section 6.12 Orthogonality of Bessel

Functions
9) Hermites differential equation with parameter is
y 2xy + y = 0
x ( , ) ,
(HE)
a) Solve the HE by the power series method

b) Consider two linearly independent solutions
y1 ( x ) = ...
y2 ( x ) = ...
which include parameter

c) If is a non-negative even integer, = 0,2,4,...,2n,... , then the series
terminates, and one obtains, alternating for y1 and y2 , polynomials of
degree n , which are called Hermitian polynomials H n ( x ) . Write them in
traditional form in which the coefficient of x n is equal to 2 n (the second

solution is not polynomial).
d) Rewrite HE in self-adjoint form and determine the weight function
w( x)
e)
Check if the HP are orthogonal with the weight function w ( x )
over ( , ) :
H m ( x ) H n ( x )w ( x ) dx = 0
if m n
f) Give an example of function representation into Fourier-Hermite series

Special Functions

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Special Functions

Hochgeladen von

Copyright:

Verfügbare Formate

Chapter 5

Chapter 5 SPECIAL FUNCTIONS

5.1 Heaviside step function

5.7 Legendre Functions

Chapter 5 SPECIAL FUNCTIONS

Chapter 5 SPECIAL FUNCTIONS

5.1 Heaviside Step Function

5.1 Heaviside Function (unit step function)

Graphically it can be shown as:

Shifting of the step function along the x-axis:

The filter function can be constructed in terms of the step function:

It cuts the values of functions to zero outside of the interval [a , b] :

The Heaviside step function is used for the modeling of a sudden

Chapter 5 SPECIAL FUNCTIONS

5.2 Dirac Function (delta function)

where a < 0 and b > 0

The delta function is vanishingly narrow at x = 0 but nevertheless

b) Gauss distribution functions:

d) Cauchy density (distribution) functions:

Chapter 5 SPECIAL FUNCTIONS

1) Extension of the interval of integration to all real numbers still keeps

2) The Dirac delta function is a generalized derivative of the Heaviside

Therefore, the step function is formally an antiderivative of the delta

Chapter 5 SPECIAL FUNCTIONS

Integration with derivatives of the delta function (integration by parts):

The delta function is applied for modeling of impulse processes.

Chapter 5 SPECIAL FUNCTIONS

5.3 Sine Integral Function

Graphically it can be shown as:

In Chapter 3, we discussed connection of Gibbs phenomena in the Fourier

Chapter 5 SPECIAL FUNCTIONS

5.4 Error Function

The complimentary error function is defined as

Series expansion of the error function:

Derivatives of the error function:

Chapter 5 SPECIAL FUNCTIONS

5.5 Gamma Function

The Gamma function appears in many integral or series representations

which for natural values p,q

Chapter 5 SPECIAL FUNCTIONS

When x = n is a natural number then

n = 0 ,1,2 ,... provided that 0! = 1

The gamma function is a generalization to real numbers of a

(1) = 1 then using property (a)

The gamma function is differentiable everywhere except at

Stirling formula (approximation for large x , x > 9 )

Calculation of gamma function: Lanczos approximation in

Chapter 5 SPECIAL FUNCTIONS

In the method of separation of variables applied to a PDE in cylindrical

Simplify equation by the change of variables to eliminate parameter :

Now the equation is written in traditional variables, and it includes only

Singular points of the differential equation with variable coefficients are

Identify coefficients of the equation in normal form:

Substitute coefficients p0 and q0 into the indicial equation (Ch. 2.42):

Chapter 5 SPECIAL FUNCTIONS

and substitute them into the Bessel Equation (1):

(n + )(n + 1)cn x n+ + (n + )cn x n+ + cn x n+ +2 2 cn x n+

Divide the equation by x and collect the terms

n(n + 2 )cn x n + cn x n+2 = 0

m(m + 2 )cm x m + cm2 x m = 0

Combine both series:

0 c0 + (1 + 2 )c1 x + [m(m + 2 )cm + cm2 ]x m = 0

(by assumption, > 0 )