Special Functions

QUASI-LINEAR 1st ORDER ODEs

Complex Quasi-Linear 1st Order ODEs

General form:
w ( z ) + a ( z ) w ( z ) = f ( z )

on a domain D with IC w ( zo ) = wo

i.e:
linear in the unknown function w ( z ) and its derivative
with variable coefficient a ( z ) and forcing (or known) term f ( z )

The associated homogeneous (or reduced) equation is obtained by setting f ( z ) = 0 :

w ( z ) + a ( z ) w ( z ) = 0

Look for an integrating factor h ( z ) such that the homogeneous equation can be written as:
1
h
( hw ) = w + w = f
h
h

h
a = = ( ln h )
h

h ( z) = e

a ( ) d + c

a ( ) d

Hence, integrating:
h(z) w(z) =

h ( ) f ( ) d + c

with IC

h ( zo ) w ( zo ) =

h ( ) f ( ) d + c
zo

Therefore the most general solution of 1st order quasi-linear equations is:
h ( zo )
1 z
w(z) =
w ( zo ) +
h ( ) f ( ) d
h(z)
h ( z ) zo
with the second term equal to zero for homogeneous equations ( f = 0 ).
Luca dAgostino, 2015/16.

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
Complex Quasi-Linear 2nd Order ODEs

General form:
w ( z ) + p ( z ) w ( z ) + q ( z ) w ( z ) = f ( z )
on a domain D
linear in the unknown function w ( z ) and its derivatives
with variable coefficients p ( z ) , q ( z ) and forcing (or known) term f ( z )

The associated homogeneous (or reduced) equation is obtained by setting f ( z ) = 0 :

w ( z ) + p ( z ) w ( z ) + q ( z ) w ( z ) = 0

A point zo in the closure of D is:

an ordinary point if p ( z ) and q ( z ) have at most a removable singularity at zo

a regular singular point if:
p ( z ) and q ( z ) have at most a pole of order 1 and 2 at zo , respectively
an irregular singular point in all other cases

Example:
w +

1
1
w
+
w=0

z2 z2 + 1
z z2 + 1

on the entire complex plane

z = 0 is a pole of order 2 for p ( z ) and 1 for q ( z ) z = 0 is an irregular singular point

z = i are regular singular points
z = i are simple poles for both p ( z ) and
Luca dAgostino, 2015/16.

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
Linearly Independent Functions

Two functions f1 ( z ) and f2 ( z ) , defined and not identically zero on a domain D , are:
linearly dependent if for some constants c1 and c2 not both zero:
c1 f1 ( z ) + c2 f2 ( z ) = 0
identically on D

linearly independent otherwise

The Wronskian W ( z )
f (z)
W f1 f2 ( z ) = 1
f1( z )

of f1 ( z ) and f2 ( z ) , analytic on a domain D , is:

f2 ( z )
= f1 ( z ) f2 ( z ) f2 ( z ) f1( z )
f2 ( z )

Clearly:
f1 ( z ) , f2 ( z ) linearly dependent

W f1 f2 ( z ) = 0

(proportional columns of W f1 f2 )

Viceversa:

W f1 f2 ( z ) = f22 ( z )

d f1 ( z )
=0
dz f2 ( z )

Hence, finally:
W f1 f2 ( z ) = 0

f1 ( z ) f2 ( z )

f1 ( z ) , f2 ( z ) linearly dependent

(necessary and sufficient condition)

Luca dAgostino, 2015/16.

(linearly dependent)

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
Properties of the Homogeneous Solutions

If w1 and w2 are two solutions of the associated homogeneous equation:

and
on D
w1 + pw1 + qw1 = 0
w2 + pw2 + qw2 = 0
Eliminating q by cross-multiplication by w1 and w2 and subtraction:
w1w2 w2 w1 + p ( w1w2 w2 w1 ) = 0 Ww1w2 + pWw1w2 = 0

where Ww1w2 ( z ) = w1w2 w2 w1 (Wronskian of w1 , w2 ), and integrating with Ww1w2 ( zo ) = Wo :

Ww1w2 ( z ) = Wo e

zz p ( ) d
o

Wo = 0 w1 , w2 linearly dependent
p d
since e
0

W

w
,
w

linearly
independent
1
2
o

Let w satisfy the homogeneous equation with homogeneous ICs:

with
w ( zo ) = 0 and w ( zo ) = 0
w + pw + qw = 0
Then, from this equation and its higher order derivatives:
w( k ) ( zo ) = 0 recursively for any order of differentiation k 2
and, by Taylor's expansion and the identity theorem:

w ( k ) ( zo )
w(z) =
( z zo )k = 0 for z zo < R
k!
k=0

w0

identically on D

Hence, for any two solutions wr and ws of the complete equation with the same ICs:
w = wr ws ; w ( zo ) = w ( zo ) = 0 wr ws on D (uniqueness of the solution)

Luca dAgostino, 2015/16.

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
Particular Solutions by Variation of Parameters

Assume a particular solution wP of w + pw + qw = f in the form:

wP ( z ) = a1 ( z ) w1 ( z ) + a2 ( z ) w2 ( z )
where w1 , w2 are linearly independent solutions of w + pw + qw = 0 and:
w p = a1w1 + a2 w2 + a1w1 + a2 w2

w p = a1w1+ a2 w2 + a1w1 + a2 w2
Substituting in w + pw + qw = f obtain:

a1w1 + a2 w2 = 0

a1w1 + a2 w2 = 0

a1w1 + a2 w2 = f
which represents a linear system for the unknowns a1 and a2 whose solution is:
w f
wf
Ww1w2 = w1w2 w1w2 0
and
where
a1 = 2
a2 = 1
Ww1w2
Ww1w2
a1 ( w1+ pw1 + qw1 ) + a2 ( w2 + pw2 + qw2 ) + a1w1 + a2 w2 = f

Hence, integrating from zo to z on any simple curve in D :

z
z
w2 ( )
w1 ( )
and
a1 ( z ) =
f ( ) d
a2 ( z ) =
f ( ) d
zo W
z
o

w1 w2 ( )
w1 w2 ( )
and finally, substituting in the original expression of wP :
z
w ( ) w2 ( z ) w2 ( ) w1 ( z )
where
wP ( z ) = G ( z, ) f ( ) d
G ( z, ) = 1
zo
Ww1w2 ( )
Luca dAgostino, 2015/16.

(kernel)

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
Complete Solution of the Non-Homogeneous IVP

From earlier results, the general solution of the Initial Value Problem (IVP):
w ( zo ) = wo
w ( z ) + p ( z ) w ( z ) + q ( z ) w ( z ) = f ( z )
with ICs

w ( zo ) = wo
is:
w ( z ) = c1w1 ( z ) + c2 w2 ( z ) + w p ( z )
with w1 , w2 linearly independent integrals of w + pw + qw = 0 and, from the ICs:

c1 , c2 solutions of:

c1w1 ( zo ) + c2 w2 ( zo ) + w p ( zo ) = wo

c1w1 ( zo ) + c2 w2 ( zo ) + wp ( zo ) = wo

The determinant of the coefficients of this system is:

Ww1w2 ( zo ) = w1 ( zo ) w2 ( zo ) w1 ( zo ) w2 ( zo ) 0
and therefore c1 , c2 are uniquely determined as:

wo w p ( zo ) w2 ( zo ) w2 ( zo ) wo wp ( zo )
c1 =
Ww1w2 ( zo )
c2 =

w1 ( zo ) wo wp ( zo ) wo w p ( zo ) w1 ( zo )

Luca dAgostino, 2015/16.

Ww1w2 ( zo )

for w1 , w2 linearly independent

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
Power Series Solutions of the Homogeneous Equation (Frobenius Method)

Expand in Taylors series:

k 1
k
w
z
=
ka
z
=
k
+
1
a
z

(
)
(
)

k
k
+1

k =1
k=0
w ( z ) = ak z k

k=0
w ( z ) = k ( k + 1) a z k 1 = ( k + 1) ( k + 2 ) a z k

k +1
k+2

k =1
k=0
Using similar expansions for p ( z ) and q ( z ) for z < R (convergence radius):

k
k
k
k
p ( z ) = pk z
p ( z ) w ( z ) = pk z ( k + 1) ak +1z = ( n + 1) an +1 pk n z

k=0
k=0
k=0
k =0 n=0

k
q ( z ) = q z k
q ( z ) w ( z ) = q z k a z k =
an qk n z k

k
k
k

k=0
k=0
k=0
k =0 n=0
and, substituting in the homogeneous ODE:

w + pw + qw = 0 ( k + 1) ( k + 2 ) ak + 2 + ( n + 1) an +1 pk n + an qk n z k = 0
k=0
n=0

from which:
k
a0 w1 ( z ) for a1 = 0
1
ak + 2 =
n
+
1
a
p
+
a
q

w
z
=
(
)
(
)

n +1 k n
n kn

( k + 1) ( k + 2 )
n=0
a1w2 ( z ) for a0 = 0

Hence the most general solution of w + pw + qw = 0 is:

wh ( z ) = a0 w1 ( z ) + a1w2 ( z )
with a0 and a1 arbitrary constants

Luca dAgostino, 2015/16.

Special Functions
QUASI-LINEAR 2nd ORDER ODEs
The Euler Equation

Consider the Euler or equidimensional equation:

r
s
w + 0 w + 02 w = 0
z
z
where:
r0 z pole of O (1) and s0 z 2 pole of O ( 2 )

Assume:
w ( z ) = z = e Log z
where:
z 2 / 0

z = 0 is a regular singular point

with constant ( 1) + r0 + s0 z 2 = 0

2 + ( r0 1) + s0 = 0

(indicial equation)

whose solutions are:

1,2 =

1 r0

Hence:
1 2

(1 r0 )2 4s0
2

with

w ( z ) = c1z1 + c2 z 2

and, by direct substitution:

w ( z ) = c1z1 + c2 z1 Log z
1 = 2
Luca dAgostino, 2015/16.

1 2

1 = 2

if (1 r0 ) 4s0 0
2

if (1 r0 ) 4s0 = 0
2

Special Functions
IMPROPER INTEGRALS
Improper Integrals of Complex Functions

Let f ( z,t ) be a complex function:

defined for complex z in S and real t in [ a, ) (respectively [ a,b ) )
continuous in t on [ a, ) (respectively [ a,b ) ) for any z of S

The improper integral of the first (respectively second) kind of f ( z,t ) :

f ( z,t ) dt = lim f ( z,t ) dt = F ( z ) respectively

a

for T > ( , z )

f ( z,t ) dt F ( z ) <
a

T b

respectively

T b < ( , z )

is uniformly convergent if, in particular, = ( ) is independent on z

is absolutely convergent if:

f ( z,t ) dt = lim f ( z,t ) dt = F ( z )

is defined and converges to F ( z ) if for any > 0 there exists a ( , z ) such that:
T

f ( z,t ) dt <

respectively

f ( z,t ) dt <
b

Weierstrass M -criterion for convergence:

f ( z,t ) M ( t )

M ( t ) dt <
a

Luca dAgostino, 2015/16.

f ( z,t ) dt uniformly and absolutely convergent on S

a

Special Functions
IMPROPER INTEGRALS
Differentiation of Integrals

A complex function f ( z,t ) on D [ a,b ] is:

jointly continuous at zo ,t o of D [ a,b ] if for any > 0 there is a ( , zo ,t o ) such that:

f ( z,t ) f ( zo ,t o ) <

z zo + t t o < ( , zo ,t o )

whenever

If f ( z,t ) is jointly continuous on D [ a,b ] and analytic on D for any t of [ a,b ] then:
F (z) =

f ( z,t ) dt
b

is also analytic on D with derivatives:

k
b f ( z,t )
(k)
F (z) =
dt
for k = 1, 2,...
k
a
z

(differentiation under the integral sign)

The same result holds for improper integrals on infinite or finite (open) intervals ( a,b ) if:
the integrand f ( z,t ) is defined and jointly continuous on D ( a,b )
the integral is uniformly convergent on every bounded and closed subset of D

Examples:
Jn ( z) =

(z) =

cos ( nt z sin t ) dt

t z 1et dt

Luca dAgostino, 2015/16.

for Re z > 0

(Bessel functions of order n )

(Gamma function)

Special Functions
THE GAMMA FUNCTION
Definition, Convergence and Analyticity

Recall Eulers integral definition of the gamma function:

(z) =

t z 1et dt

and also

(z) =

t z 1et dt + t z 1et dt
1

where:

t z 1 = eln t

z1

x1

The integrand t z 1et :

is jointly continuous for any z and t > 0
for t > 1 satisfies the Weierstrass M -criterion for convergence:
t x t
z 1 t
t e = e t et M ( t ) on Re z
t
for 0 < t 1 , using Taylors expansion of et and the ratio test of uniform convergence:

e =

z 1 t

( 1)k t k + z 1

k=0

= e( x +iy1) ln t = e( x 1) ln t +iy ln t = eln t ei ln t = t x 1 cos ln t y + i sin ln t y

k!t k + z
k!t k + x
t
lim
=
lim
=
lim
=0
k ( k + 1)!t k + z 1
k ( k + 1)!t k + x 1
k k + 1

k!

Hence the series for t z 1et can be termwise integrated and:

(z) =

t
0

e dt + t

z 1 t

z 1 t

1)
(
dt =
+ t z 1et dt
1
k = 0 k!( k + z )

is uniformly convergent and analytic for all z 0, 1, 2,... , where it has simple poles
Luca dAgostino, 2015/16.

Special Functions
THE GAMMA FUNCTION
Properties of the Gamma Function

From the definition:

(z) =

t z 1et dt

(1) = et dt = et = 1
0
0

2
(1 2 ) = t 1 2 et dt = 2 eu du = 2
0
0

2 =

Integrating by parts:

( z + 1) =

t z et dt = t z et 0 + z t z 1et dt = z ( z )
0

and in particular, with z = 1 and it follows by induction:

( n + 1) = n!
for n = 0,1, 2... (1) = 1 = 0!

Besides:

( z ) (1 z ) =

t z 1et dt

z e d =

With the change of variables:

u
uv
and
t=
>0
=
>0
1+ v
1+ v

dt d t z 1 z et
0

( ,t ) u v
u
=
=
0
( u, v ) tu t v (1 + v )2

obtain by real integration in u and residue integration in v :

z 1

u
( z ) (1 z ) = e du
dv = 1
=
0
0 1+ v
sin z sin z
Luca dAgostino, 2015/16.

Special Functions
THE GAMMA FUNCTION
The Logarithmic Derivative of the Gamma Function

Define:

(z)
(logarithmic derivative of ( z ) )
(z)
By logarithmic differentiation of ( z + 1) = z ( z ) :
1
( z + 1) = ( z ) +
z
In particular, recursive use of this equation for integer values of z = k yields:
1
1
( k + 1) = +
+ ... + 1 + (1)
k k 1
where:

(1)
(Euler constant)
(1) =
= (1) = et Log t dt = = 0.57721566...
0
(1)
(z) =

Next, by logarithmic differentiation of:

( z ) (1 z ) =
sin z
obtain:

( z ) (1 z ) =
Luca dAgostino, 2015/16.

cos z
= cot z
sin z

Special Functions
BESSEL FUNCTIONS
Laplaces Equation in Cylindrical Coordinates

Laplaces equation in orthogonal cylindrical coordinates r , , z :

2u 1 u 1 2u 2u
2
u= 2 +
+
+
=0
r
r r r 2 2 z 2
Assume a separable solution of the form:
R
R

u = R ( r ) ( ) Z ( z ) R + + 2 Z + R Z = 0
r
r

Separating the terms that depend on z :

Z
1
R
= R + 2 = 2 = constant
Z
R
r r
since both terms are functions of different independent variables, and again:

r2
R
= R + 2 r 2 = 2 = constant

R
r

Hence the original 3D PDE problem for u is transformed in three ODEs for R , and Z :
R 2 2
Bessel equation for R ( r )
R +
+ 2R=0
r
r

+ 2 = 0
Z 2 Z = 0
Luca dAgostino, 2015/16.

harmonic equation for ( )

hyperbolic equation for Z ( z )

Special Functions
BESSEL FUNCTIONS
Bessel Functions of the 1st Kind

Introducing the transformation:

z = r
d
d

dr
dz
w ( z ) = R ( r )

in

the ODE for R ( r ) reduces to:

2
1
w + w + 1 2 w = 0
z
z

(Bessel equation)

(Bessel equation of order in standard form)

Using Frobenius method, the solutions of Bessels equation are:

k
2 k
1)
(
z
J ( z ) =

k = 0 k! ( k + 1) 2

R 2 2
R +
+ 2R=0
r
r

(Bessel functions of the 1st kind)

From the general expression of the Wronskisan WJ , J e p ( z ) dz and the properties of ( z ) :

zWJ , J ( z ) = ze dz z = A = lim zWJ , J ( z ) =

z0

2
2 sin
=
(1 + ) (1 )

Hence, for n = 0, 1, 2,... (integer):

WJ , J ( z ) 0 J ( z ) , J ( z ) linearly independent

Luca dAgostino, 2015/16.

Special Functions
BESSEL FUNCTIONS
Integral Representation

Using the Taylor expansions of the exponential function:

1 z ( 1) z
( 1) z
G ( z, ) = e
=e e =
=

2 k = 0 k! 2
m = 0 m!
m = 0 k = 0 m!k! 2
Eliminating m = n + k and using the definition of J n ( z ) :
z ( 1 ) 2

z
2

z
2

k
2k+n

1) z
(
n
G ( z, ) =
= J n ( z ) n

n = k = 0 ( k + n )!k! 2
n =

m+k

mk

(Laurents expansion of G )

Hence, for any simple contour C like = ei not through = 0 where G is singular:

Jn ( z) =

G ( z, )
1
1 z ( ei e i ) 2 i ( n +1) i
d

=
e
e
ie d
n +1

2 i

2 i

or, using the definition of sin :

1 i ( z sin n )
1
Jn ( z) =
e
d

=
2
2

cos ( z sin n ) + i sin ( z sin n ) d

and finally:
1
(integral representation of J n ( z ) )
cos ( z sin n ) d
0
since cos ( z sin n ) and sin ( z sin n ) are respectively even and odd functions
Jn ( z) =

Luca dAgostino, 2015/16.

Special Functions
BESSEL FUNCTIONS
Bessel Functions of the 2nd Kind

Define:

J ( z ) cos J ( z )
for n

sin
Y ( z ) =
Y ( z ) = lim Y ( z )
for = n

n
n

(Weber's Bessel function of the 2nd kind)

For = n , using l'Hpital rule:

1 J (z)
n J ( z )
Yn ( z ) =
( 1)

n

where, from the definitions of J ( z ) and the logarithmic derivative ( z ) of ( z ) :

J ( z )
( 1)
z
=

k = 0 k! ( k + 1) 2
k

2 k

Log

+
1
(
)

(log-singular at z = 0 )

Since J n ( z ) and Yn ( z ) are linearly independent by construction:

w ( z ) = c1 J ( z ) + c2Y ( z )

with c1 and c2 arbitrary constants

is the general solution of Bessel's equation of any order (integer or not)

Notice that, if the principal value z is considered, J ( z ) and Y ( z ) are:

analytic on the cut z -plane z > 0 , < < , for n 0 , entire for = n 0

Luca dAgostino, 2015/16.

Special Functions
BESSEL FUNCTIONS
Modified Bessel Functions

With the transformation = iz and w ( ) = w ( iz ) , from the Bessel equation obtain:

2
1
w* + w* 1 + 2 w* = 0

(modified Bessel equation of order )

Since the solutions J ( iz ) and Y ( iz ) fail to be real-valued when z = x > 0 , define:

(modified Bessel function of the 1st kind)
I ( z ) = ei 2 J ( iz )
K ( z ) = i ( 2 ) ei 2 J ( iz ) + iY ( iz )

(modified Bessel function of the 2nd kind)

Recalling the definitions of J ( z ) , Y ( z ) and using i 2 k i = ( 1) e Log i = ( 1) ei 2 :

k

2k
( 1)k
z
k! ( k + + 1) 2
k=0
J ( z ) cos J ( z )
=

z
J ( z ) =
2
Y ( z )

sin

1
z
I ( z ) =

k = 0 k! ( k + + 1) 2
K ( z ) =

2 k +

I ( z ) I ( z )
2 sin

Since I ( z ) and K ( z ) are linearly independent by construction:

w ( z ) = c1 I ( z ) + c2 K ( z )
with c1 and c2 arbitrary constants
is the general solution of the modified Bessel's equation of any order (integer or not)

Notice that, if the principal value z = e Log z is considered, I ( z ) and K ( z ) are:

analytic on the cut z -plane z > 0 , < < , for n 0 , entire for = n 0

Luca dAgostino, 2015/16.

Special Functions
BESSEL FUNCTIONS
Closed-Form and Asymptotic Representations

With the position w ( z ) = z 1 2u ( z ) :

2
2 1 4
1
w + w + 1 2 w = 0 u + 1
u=0
z
z
z 2

Hence for = 1 2 or z Bessel's equation reduces to the harmonic equation:

u + u = 0

and

J ( z )
A cos z + Bsin z
= z 1 2u
Y ( z )
z1 2

A rather elaborate analysis shows that for z :

2

cos z ( 2 + 1)
z
4

and for = 1 2 :
J ( z )

J1 2 ( z ) =

2
sin z
z

and

and

Y1 2 ( z ) =

Y ( z )

sin z ( 2 + 1)
z
4

2
cos z
z

Similarly, for the modified Bessel equation:

ez
I ( z )
= I1 2 ( z )
2 z
(independent on )
Luca dAgostino, 2015/16.

and

e z
K ( z )
= K1 2 ( z )
2z

z
for
= 1 2

Special Functions
BESSEL FUNCTIONS
Some Useful Properties

From the relevant series definitions:

B
z
+
B
z
=
2
B ( z )
(
)
(
)
1
+1
z

B 1 ( z ) B +1 ( z ) = 2 B ( z )

B ( z ) B ( z )
z

I 1 ( z ) I +1 ( z ) = 2 I ( z )
z

I 1 ( z ) + I +1 ( z ) = 2 I ( z )

I 1 ( z ) = I ( z )

K
z

K
z
=
2
K ( z )
(
)
(
)
1
+1
z

K 1 ( z ) + K +1 ( z ) = 2 K ( z )

K 1 ( z ) = K ( z ) K ( z )
z

and in particular for = 0 :

Z 0 ( z ) = Z1 ( z ) ;
I 0 ( z ) = I1 ( z )

B 1 ( z ) =

where B = J, Y

I ( z )
z

K 0 ( z ) = K1 ( z )

and

With the same procedure:

z B ( z ) = z B 1 ( z )

z K ( z ) = z K 1 ( z )

Luca dAgostino, 2015/16.

( B = J, Y , I )

z B ( z ) = z B +1 ( z )

z I ( z ) = z I +1 ( z )

( B = J, Y , K )

Special Functions
BESSEL FUNCTIONS
Transformed Equations and Solutions

The solution of Bessels equation:

2
1
w + w + 1 2 w = 0
z
z

is:

w ( z ) = c1 J ( z ) + c2Y ( z )

With the transformation:

d s
z=

and

u ( ) =

1 a
b
r
2
r

d s
w

s

the Bessel equation becomes:

1
1
u + a + 2b r u + 2 c + d 2 s b (1 a r ) r + b 2 2r u = 0

and its solution is:

u ( ) =

1 a
b
r
2
r

where:

1 1 a
=

c
s
2
2

Luca dAgostino, 2015/16.

c1 J s d s + c2Y s d s

Special Functions
LEGENDRE FUNCTIONS
Laplaces Equation in Spherical Coordinates

Laplaces equation in orthogonal spherical coordinates r , , :

1 2 u
1
u
1
2u

2
u= 2
sin + 2 2
=0
r
+

r sin 2
r r r r 2 sin
Assume a separable solution of the form:
r 2 R r
( sin ) R
u = R ( r ) ( ) ( )

+
R

+ 2 2 =0
r2
r 2 sin
r sin
Separating the terms that depend on :

( r R )r sin2 ( sin ) sin = m 2 constant

=

since both terms are functions of different independent variables, and again:
( r 2 R )r = m 2 ( sin ) = n ( n + 1) constant
R
sin 2
sin
2

Hence the original 3D PDE problem for u is transformed in three ODEs for R , , :
equidimensional equation for R
r 2 R + 2rR n ( n + 1) R = 0
+ cot + n ( n + 1) m 2 sin 2 = 0
Legendre-type equation for
+ m 2 = 0
harmonic equation for

Luca dAgostino, 2015/16.

Special Functions
LEGENDRE FUNCTIONS
Radial Equation

Recall:
r 2 R + 2rR n ( n + 1) R = 0
Assuming a solution in the form R = r obtain:
n
( + 1) n ( n + 1) = 0 (indicial equation) 1,2 =
1 n

and therefore:
Rn ( r ) = c1r n + c2 r 1 n
Azimuthal Equation

Recall:

+ m 2 = 0
Assuming a solution in the form = e obtain:
2 + m 2 = 0 (associated algebraic equation) = im
and therefore:
m ( ) = c1eim + c2 eim

( 2 m -periodic)

For physically meaningful (real periodic) solutions m = 0, 1, 2, ... integer

Luca dAgostino, 2015/16.

Special Functions
LEGENDRE FUNCTIONS
Legendre Equation and Polynomials

Recall:

m2
+ cot + n ( n + 1) 2 = 0
sin

With the transformation:

z = cos

w
z
=

(
)
(
)

( ) = w ( z ) sin

2
( ) = w ( z ) sin w ( z ) cos

the above equation reduces to the associated Legendre equation of degree n and order m :

m2
2
1 z w 2zw + n ( n + 1)
w=0
2
1

Using Frobenius method, a solution of the above equation with n = 0, 1, 2, ... and m = 0 is:
n
1 dn 2
Pn ( z ) = n
z

1
(Rodrigues formula for Legendre polynomials Pn ( z ) )
2 n! dz n

Some relevant properties:

1
2
2
P
x
dx
=
(
)
1 n
2n + 1
( n + 1) Pn +1 ( z ) = ( 2n + 1) zPn ( z ) nPn 1 ( z )

Luca dAgostino, 2015/16.

(norm)
(recursive relation)

Special Functions
LEGENDRE FUNCTIONS
Associated Legendre Equation

Recall:

m2
1 z w 2zw + n ( n + 1)
w=0
2
1

The general solution of this equation for n = 0, 1, 2, ... , m = 0 and z = x real is:
w ( x ) = c1Pn ( x ) + c2Qn ( x )
for x < 1
where Qn ( x ) is the Legendre function of second kind of order n :

n
1 dn 2
1+ x 1
1+ x
Qn ( x ) = n
x

1
log

P
x
log
for x < 1
(
)
n
2 n! dx n
1 x 2
1 x
The solution for z complex is obtained by analytic continuation on the z -plane:
w ( z ) = c1Pn ( z ) + c2Qn ( z )
for any z
by properly choosing the logarithmic branch cuts Qn ( z ) = Qn ( x ) on 1 < x < 1

Finally, it can readily be shown by direct substitution that:

m
m
m2 d
m2 d
2
2
w ( z ) = c1 z 1
Pn ( z ) + c2 z 1
Qn ( z )
dz m
dz m
is the general solution of the associated Legendre equation with n = 0, 1, 2, ... and m 0

Luca dAgostino, 2015/16.

Special Functions
BOUNDARY VALUE PROBLEMS
2nd Order Sturn-Liouville BVPs

A linear homogeneous ODE for y ( x ) on ( x1 , x2 ) :

y + py + ( q + r ) y = 0 y ( x ) = A ( x, ) + B ( x, )
with:
real coefficients p ( x ) , q ( x ) and r ( x )
constant free parameter (yet undetermined)
and either type of BCs at x = xi = x1 , x2 :
with ai , bi constants, not both zero
ai y ( xi ) + bi y ( xi ) = 0
y ( x1 ) = y ( x2 )

and

y ( x1 ) = y ( x2 )

(general solution)

(homogeneous BCs)
(periodic BCs)

Then, since either type of BCs is linear and homogeneous in A and B :

Ac11 ( ) + Bc12 ( ) = 0
A = B = 0 y ( x ) = 0 (trivial solution)

Ac21 ( ) + Bc22 ( ) = 0
except for values of such that:

cij ( ) = c11 ( ) c22 ( ) c12 ( ) c21 ( ) = 0 = h , h = 1, 2, 3, ... (eigenvalues)

Then the equations for A and B are lineraly dependent and, using for instance the 1st one:
c ( )
(eigenfunctions)
B = A 11
yh ( x ) c12 ( h ) ( x, h ) c11 ( h ) ( x, h )
c12 ( )
Notice that the eigenfunctions yh ( x ) are only determined to within a multiplicative factor

Luca dAgostino, 2015/16.

Special Functions
BOUNDARY VALUE PROBLEMS
Orthogonality of the Eigenfunctions

Consider a Sturm-Liouville BVP on ( x1 , x2 ) for the ODE:

ai y ( xi ) + bi y ( xi ) = 0 at x = xi = x1 , x2
y + py + ( q + r ) y = 0
with either of

y ( x1 ) = y ( x2 ) and y ( x1 ) = y ( x2 )
With the transformations:
p ( x ) dx
;
v ( x ) = qu
and
w ( x ) = ru
(all real)
u ( x ) = e
the above ODE becomes:
(uy ) + ( v + w ) y = 0

For any two eigenvalues h , k and eigenfunctions yh ( x ) , yk ( x ) , by complex conjugation:

(uyh ) + vyh = h wyh and (uyk ) + vyk = k wyk
By cross-multiplication by yk and yh , subtraction to eliminate v , and integration by parts:

h k

x2
x1

x2
x2
x2
wyh yk dx = yh ( uyk ) dx yk ( uyh ) dx = yh ( uyk ) yk ( uyh ) x
x1

x1

Then, using the BCs (either homogeneous or periodic) at x = x1 , x2 :

x1
x2

=
u
y
y

y
y

(
)
h k
k h x = 0 (orthogonality relations)
x2
2
1
h = k i2 Im {h } yh w dx =
x1

Hence, distinct eigenfunctions are orthogonal on ( x1 , x2 ) w.r.t. the weight function w ( x )

hk

h k

Luca dAgostino, 2015/16.

x2

wyh yk dx =

Special Functions
BOUNDARY VALUE PROBLEMS
Proper 2nd Order Sturn-Liouville BVPs

A Sturm-Liouville BVP for ( uy ) + ( v + w ) y = 0 on ( x1 , x2 ) is proper if:

u > 0;
v0
w > 0 on ( x1 , x2 )
and

b1 a1 < 0 at x = x1 and

b2 a 2 > 0 at x = x2

Then, from the orthogonality relation with h = k :

h h

x2
x1

yh yh w dx = i2 Im {h }

x2
x1

yh w dx = 0 h real yh ( x ) real
2

Also, if yh ( x ) , yk ( x ) have equal eigenvalues h = k , integrating and using the BCs:

yh ( uyk ) yk ( uyh ) = ( h k ) yh yk w = 0 u ( yh yk yk yh ) = u ( yh yk yk yh ) x = 0
i

Hence:
yh ( x ) and yk ( x ) are linearly dependent
the linearly independent eigenfunction corresponding to each eigenvalue is unique

Finally, multiplying the equation ( uyh ) + vyh = wh yh by yh and integrating:

h yh yh w dx = [ uyh yh ]x2 + uyh yh dx + ( v ) yh yh dx > 0 h 0

1
x1
x1
x1

x2

>0

and, in particular:
if and only if
h = 0
Luca dAgostino, 2015/16.

x2

x2

yh constant and v = 0 identically on ( x1 , x2 )

Special Functions
BOUNDARY VALUE PROBLEMS
The BVP for the Harmonic Equation with Periodic BCs

Consider the Sturm-Liouville problem for:

y + 2 y = 0

on x1 x x2

with

y ( x1 ) = y ( x2 )

y ( x1 ) = y ( x2 )

Writing the harmonic equation as:

u = 1, v = 0, w = 1 (proper BVP) h = h2 0 h real
2

( y ) + wy = 0
y ( x ) = A cos x + Bsin x
Then, from the BCs:
A
a1 cos x1 b1 sin x1 a1 cos x1 b1 sin x1 A

=
c

a cos x b sin x a sin x + b cos x B ij ( ) B = 0

2
2
2
2
2
2
2
2

cij ( ) = 0 h2 with h = 1, 2, 3, ...

yh ( x ) = ( sin h x1 sin h x2 ) cos h x ( cos h x1 cos h x2 ) sin h x

Besides:

x2
x1

yh ( x ) yk ( x ) dx = 0

for h k

by the orthogonality relations, and for h = k can be computed using:

cos 2 h x
cos 2 h x
1
sin 2 h x
sin 2 h x dx = 2 x 4 h + c and sin h x cos h x dx = 2 h + c
Luca dAgostino, 2015/16.

Special Functions
BOUNDARY VALUE PROBLEMS
The BVP for the Harmonic Equation with Homogeneous BCs

Consider the Sturm-Liouville problem for:

y + 2 y = 0

on x1 x x2

with

a1 y ( x1 ) + b1 y ( x1 ) = 0

a2 y ( x2 ) + b2 y ( x2 ) = 0

Writing the harmonic equation as:

u = 1, v = 0, w = 1 (proper BVP) h = h2 0 h real
2

( y ) + wy = 0
y ( x ) = A cos x + Bsin x
Then, from the BCs:
A
a1 cos x1 b1 sin x1 a1 cos x1 b1 sin x1 A

=
c

a cos x b sin x a sin x + b cos x B ij ( ) B = 0

2
2
2
2
2
2
2
2

cij ( ) = 0 h2 with h = 1, 2, 3, ...

yh ( x ) = ( a1 sin h x1 + b1 h cos h x1 ) cos h x ( a1 cos h x1 b1 h sin h x1 ) sin h x

Besides:

x2
x1

yh ( x ) yk ( x ) dx = 0

for h k

by the orthogonality relations, and for h = k can be computed using:

cos 2 h x
cos 2 h x
1
sin 2 h x
sin 2 h x dx = 2 x 4 h + c and sin h x cos h x dx = 2 h + c
Luca dAgostino, 2015/16.

Special Functions
BOUNDARY VALUE PROBLEMS
The BVP for the Bessel Equation in a Cylindrical Region

Consider the Sturm-Liouville problem for:

2 2
1
y + y + 2 y = 0
x
x

on 0 x l

with

yy = finite as x 0

ay + by = 0 at x = l

and 2 real. Writing the Bessel equation as:

2
2
2

u
=
x,
v
=

x,
w
=
x

0 h real

h
h
2
( xy ) + x y = 0
x

y ( x ) = AJ ( x ) + BY ( x )

Then, from the BCs:

Y ( z ) as z 0
aJ ( l ) + b J ( l ) = 0

B=0

y ( x ) = AJ ( x ) = J ( x )

l = h l = h

h2 = h2 l 2 = h

yh ( x ) J ( h x )

h = 1, 2, 3, ...

Besides:

J ( x ) J ( x ) x dx = 0
l

for h k

by the orthogonality relations, and for any value of h = k can be computed using:
1 2 2
2
J

x
x
dx
=
x J ( h x ) J 1 ( h x ) J +1 ( h x ) + c
(
)
h
2
Luca dAgostino, 2015/16.

Special Functions
BOUNDARY VALUE PROBLEMS
The BVP for the Bessel Equation in an Annular Region

Consider the Sturm-Liouville problem for:

2 2
1
y + y + 2 y = 0
x
x

on x1 x x2

a1 y ( x1 ) + b1 y ( x1 ) = 0

a2 y ( x2 ) + b2 y ( x2 ) = 0

with

and 2 real. Writing the Bessel equation as:

u = x, v = 2 x, w = x h = h2 0 h real
2
2

( xy ) + x y = 0
x

y ( x ) = AJ ( x ) + BY ( x )

Then, from the BCs:

a1 J ( x1 ) + b1 J ( x1 ) a1Y ( x1 ) + b1Y ( x1 ) A
A

=
c

(
)

ij
B = 0
B
a
J

x
+
b

x
a
Y

x
+
b

(
)
(
)
(
)
(
)
2
2

2
2
2
2

2
2
cij ( ) = 0 h2 = h with h = 1, 2, 3, ...

yh ( x ) = a1Y ( h x1 ) + b1Y ( h x1 ) J ( h x ) a1 J ( h x1 ) + b1 J ( h x1 ) Y ( h x )

Besides:

x2
x1

yh ( x ) yk ( x ) x dx = 0 for h k (by the orthogonality relations)

and for h = k can be computed either numerically or using the closed form representations of:

J ( x ) x dx ,
2

Luca dAgostino, 2015/16.

Y ( x ) x dx
2

and

J ( x )Y ( x ) x dx

Special Functions
BOUNDARY VALUE PROBLEMS
The BVP for the Legendre Equation

Consider the Sturm-Liouville problem for:

h ( h + 1)
2x
on 1 x 1
y
y
+
y=0

1 x2
1 x2
with BCs:
yy = finite at x = 1
Writing the Legendre equation as:
u = 1 x 2

2
1 x y + y = 0

v = 0

and

Here y ( x ) = APh ( x ) + BQh ( x ) and, from the BCs:

w = 1

= h ( h + 1)

Qh ( x ) as x 1 B = 0 y ( x ) = yh ( x ) = Ah Ph ( x )

The orthogonality relations for the yh s are:

Ph ( x ) Pk ( x ) dx = 0

for h k

A closed-form expression can be obtained for:

1
2
2
for any value of h
P
x
dx
=
(
)
h
1
2h + 1

Luca dAgostino, 2015/16.

(proper BVP, h 0 )

Special Functions
BOUNDARY VALUE PROBLEMS
Nonhomogeneous Sturm-Liouville BVPs

A linear nonhomogeneous ODE for y ( x ) :

y + py + ( q + r ) y = f

on

( x1, x2 )

with

real coefficients p ( x ) , q ( x ) , r ( x )

free parameter
known term f ( x )

and either type of BCs at xi = x1 , x2 :

ai y ( xi ) + bi y ( xi ) = 0
y ( x1 ) = y ( x2 )

and

with ai , bi constants, not both zero (homogeneous BCs)

y ( x1 ) = y ( x2 )

With the usual transformation:

p ( x ) dx
v ( x ) = qu ;
u ( x ) = e
;

w ( x ) = ru

(periodic BCs)

and

g ( x ) = fu = fe

the ODE becomes:

(uy ) + ( v + w ) y = g

Proper nonhomogeneous Sturm-Liouville BVPs defined (as before) when:

u > 0;

v 0 and

b1 a1 < 0 at x = x1
Luca dAgostino, 2015/16.

w > 0 on ( x1 , x2 )
and

b2 a 2 > 0 at x = x2

p ( x ) dx

Special Functions
BOUNDARY VALUE PROBLEMS
Solution of Nonhomogeneous Sturm-Liouville BVPs

Assume a solution of:

(uy ) + ( v + w ) y = g
as a series expansion in the eigenfunctions yh of the homogeneous BVP:

on ( x1 , x2 )

y ( x ) = ah yh ( x )
h =1

with unknown coefficients ah

Substituting in the nonhomogeneous differential equation and using:

(uyh ) + ( v + h w ) yh = 0 ah (uyh ) + v + h w + ( h ) w yh = g
h =1

or:

a ( ) wy
h =1

=g

Compute the coefficients ah ( h ) using the orthogonality conditions:

x2
x1

wyh yk dx = 0 for h k ak ( k ) wyk yk dx =

x2

x1

x2
x1

gyk dx

Therefore, finally:

x2

gyh dx

h
y
x

(
)

h
x2
h =1 ( )
= h
h wyh yh dx

y( x) =

x1

x1

Luca dAgostino, 2015/16.

unique solution
no solution, unless g yh