Legendre Polynomials - Lecture 8

1 Introduction
In spherical coordinates the separation of variables for the function of the polar angle results
in Legendres equation when the solution is independent of the azimuthal angle.
2
(1 x2 ) d P2 2x dP + l(l + 1)P = 0
dx dx
This equation has x = cos() with solutions Pl (x). As previously demonstrated, a series
solution can be obtained using the form;

an xn+s
P
P (x) =

Taking the derivatives, substituting into the ode, and collecting the coefficient of the same
power in x, one obtains the recursion relation for the coefficients.

(n + s)(n + s + 1) l(l + 1)
an+2 = an
(n + s + 2)(n + s + 1)
The indicial equation must also be satisfied by selection of the initial coefficient and/or start-
ing power of x. Thus

a0 s(s 1) = 0

a1 (1 + s)s = 0

We then have the choice of a1 = 0 and s = 0 or a0 = 0 and s = 0. Other choices are

incorporated in these two. Thus if we choose a1 = 0 and s = 0, then;

n(n + 1) l(l + 1)
an+2 = an
(n + 2)(n + 1)
Only even powers of x are included in the series. If a0 and s = 0 then only odd powers are
incorporated in the series and the recursion relation remains the same. For the series to be
convergent for |x| = 1 it must terminate, again as previously pointed out. The series then
takes the form;
l/2
(2l 2n)!
(1)n xl2k
P
Pl (x) = l
n=0 2 n!(l n)!(l 2n)!
This series results in a set of polynomials which may be obtained from the generating func-
tion;

1

g(x, t) = (1 2xt + t2 )1/2 = Pl (x) tl
P
l=0

Some initial polynomials are;

P0 = 1

P1 = x

P2 = (1/2)[3x2 1]

2 Orthogonality
Consider the orthogonality integral for the Legendre functions. This follows from the general
Sturm-Liouville problem. Put Legendres equation in self adjoint form;

d [(1 x2 ) dPl (x) ] + l(l + 1)P (x) = 0

l
dx dx
Then look at the equation for Pn (x) and subtract the equations for Pl and Pn after multipli-
cation of the first by Pn and the later by Pl . Integrate the result between 1. This results
in

R1
[(1 x2 )Pn Pl (1 x2 )Pl Pn ]11 + [l(l + 1) n(n + 1)] dxPl Pn = 0
1
The first term vanishes and thus the polynomials are orthogonal but not normalized to 1.
The normalization integral takes the form;

R1
N2 = dx Pn Pm
1

Use the Rodgiques formula to obtain;

Pn (x) = 1 ( d )n (x2 1)n

2n n! dx
(1)m+n R1 m m
N2 = m+n dx [( d m )(1 x2 )m ][( d n )(1 x2 )n ]
2 m!n! 1 dx dx
Integrate by parts m times noting that the surface terms vanish at 1.

(1)nm R1 nm
N2 = n+m dx ( d nm )(1 x2 )n
2 m!n! 1 dx

2
This vanishes for n > m (reverse n, m if n < m). If n = m then;

1 R1
N2 = 2n 2 dx (1 x2 )n = 2n 2+ 1
2 (n!) 1
The orthogonality integral is for the associated Legendre polynomials is expressed as;

R1 (j + m)!
dx Prm (j)Pkm (x) = 2j 2+ 1
1
(j m)!
The normailzation for the Legendre polynomial Prm is found for m = 0.

3 Recurrence Relations
The recurrence relations between the Legendre polynomials can be obtained from the gen-
erating function. The most important recurrence relation is;

(2n + 1)xPn (x) = (n + 1)Pn+1 (x) + nPn1 (x)

To generate higher order polynomials, one begins with P0 (x) = 1 and P1 (x) = x. The gen-
erating function also gives the recursion relation for the derivative.

Pn+1 (x) = (n + 1)Pn (x) + xPn (x)

4 Separation of variables in spherical coordinates

We have previously looked at separation of variables in spherical coordinates, particularly
with respect to the radial component. Now consider the component involving the polar angle
with x = cos() when the aximuthal angle id included in the separation.
2 2
(1 x2 ) d d
2 2x dx + [n(n + 1)
m ] = 0
dx (1 x2 )
The above equation is the associated Legendre equation. For the solution to remain finite at
x = 1 with n integral. The equation may be obtained from the ordinary Legendre equation
by differentiation.

dmm [(1 x2 )P (x) 2xP (x) + n(n + 1)P (x)]

n n n
dx
Thus we find;

3
 dm Pn (x)
Pnm (x) = (1 x2 )m/2
dxm
Then;

(n m)! m
Pnm (x) = (1)m P (x)
(n + m)! n
The generating function is difficult to use;

(2m)!(1 x2 )m/2
m
(x) ts
P
= Ps+m
2m m!(1 2tx + t2 )m+1/2 s=0

The recursion relation is;

(2n + 1)xPnm (x) = (n + m)Pn1

m m
(x) + (n m + 1)Pn+1 (x)

Finally;

Pnm (x) = (1)n+m Pnm (x)

Pnm (1) = 0

5 Spherical harmonics
In spherical coordinates Laplaces equation has the form;

2 V = (1/r 2) (r 2 V ) + 2 1 (sin() V ) + 1 2V
r r r sin() r sin () 2
2 2

Now identify an operator on the angles, L, such that;

(1/r 2 )L2 V = 1 (sin() V ) + 1 2V

2
r sin() r sin () 2
2 2

Then when Laplaces equation is solved by separation of variables;

L2 V = l(l + 1)V

Thus identify V = Rl (r) Ylm (, ) where Ylm (, ) is an angular eigenfunction with eigenval-
ues, l, and m. We already know that this function has the form;

Ylm (, ) Plm () eim

4
Now we make this function not only orthogonal, which it must be when integrating over and
, but also normalized to 1. Therefore integrating over the solid angle, d = sin() d d ;

R R2
nm lk = sin() d d [Ylm Ykm ]
0 0

Then;
r
(2l + 1)(l m)! m
Ylm (, ) = Pl (, ) eim
4 (l + m)!
The first few functions, YlM , which are the spherical harmonics;
p
Y00 = 1/(4)
p
Y11 = 3/(8) sin() ei

1
p
Y1 = 3/(8) sin() ei
p
Y01 = 3/(4) cos()

Note that Ylm = Ylm . The functions, Ylm (, ), are the spherical harmonics, and we
will later identify the operator, L, as proportional to the angular momentum operator in
Quantum Mechanics.

6 Second solution
We have found a series solutions, Pl (x) with x = cos(), which contain only odd powers and
one containing only even powers. But while these are linearly independent, they are not the
two linearly independent solutions solutions expected from a 2nd order ode. In fact, we have
demonstrated that a second solution must diverge at the singular points. Suppose we choose
to allow divergence by not letting the series terminate. Thus write;

= Pl + Ql

In the above, Ql will be the second solution. For the series solution we look at the form;

bn xn
P
= Pl +
n=1

Now l is chosen so that the series in even powers terminates. We then start bn with n = 1
so this series forms even powers of x which cannot terminate because l is already chosen to

5
 Figure 1: A representative example of Legendre functions of the first kind

make the even series terminate. The recursion relation is;

n(n + 1) l(l + 1)
bn+2 = b n odd
(n + 2+)(n + 1) n
We can also generate a series in even powers in the same way. This series represents the
second solution to Legendres equation and is written, Ql . Ql has singular points at = 0, .
Associated functions, as with the function, Pl , are defined and written, Qm
l (). The first few
functional forms are;

Q0 = (1/2) ln[ 11
+ x]
x
Q1 = (x/2) ln[ 11
+ x] 1
x
2
Q2 = ( 3x 4 1 ) ln[ 11
+ x ] 3x/2
x
Figure 1 shows the behavior of the first few Legendre functions of the first kind, while Figure
2 shows the behavior of the Legendre functions of the second kind.

6
 Figure 2: A representative example of Legendre functions of the first kind

7 Greens function in spherical coordinates

Now return to the expression of the Greens function in spherical coordinates. This was
developed several times previously but we are now in position to finalize this result. We
assume that;

2 G = 4(~r ~r )

Now expand the function using the complete set of spherical harmonics. Thus;

( )( ) = a Y ( )
P

Multiply by Ylm (, ) and integrate over the solid angle d. Ortho-normality projects out
the coefficients, alm = Ylm ( , ). Then the Greens function takes the form;

gl Ylm (, ) Ylm ( , )
P
G =
lm

The delta function is;

(~r ~r ) = 4(r r ) Ylm (, ) Ylm ( , )

P
lm

Substitution into Laplacess equation yields;

7
 2
r d 2 [rg] l(l + 1)g = 4(r r )
dr
This is solved by requiring the solution to be continuous and the derivative to match the
discontinuity at r = r . The two solutions to the homogeneous equation are r l and r (l+1) .
The result is;

rl

r < r
g(r, r ) = 4 r (l+1)
r l

2l + 1
r > r
r (l+1)
The Greens function is then;
l
G =
P 4 r< Y m (, )Y m ( , )
lm
2l + 1 r>l+1 l l

Alternatively expand the factor, 1

|~r ~r |
Choose r > r and

1 = (1/r) 1
|~r ~r | 1 (r /r) cos() + (r /r)2
1 P r
= l+1 Pl (cos())
|~r ~r | m r

When comparing this to the expression for the Greens function above, it automatically gives
the addition theorem.

Pl (cos()) =
P 4 m m
m
2m + 1 Yl (, )Yl ( , )
Return to the expansion of the vector form 1 . Figure 3 shows the geometry.
|~r ~r |
Apply a power series expansion in powers of r /r when r > r .

1 1
R = r[1 (r /r)cos() + (r /r)2 ]1/2
1 2 2
R = (1/r) + (r /r)cos() + (r /r) [1/2(3cos () 1)] +
1 = P (r l l/r l+1 ) P (cos())
l
R l
1 = P f m (r, r ) Y m (r )Ylm (r )
R lm
l, l

8
 z R

r
r

y
x

Figure 3: Geometry of the expansion for 1

|~r ~r |

Potential Point
z
q
R
a
2q
y

a
q
x

Figure 4: Geometry for the potential problem

8 Example1
From the figure note that the potential can be represented by;
 
1 2q q q
V = 4 + +
~
|R| ~ ~a|
|R ~ + ~a|
|R
Then by expansion in Legendre polynomials where R > a;

1 P al
= Pl (cos())
~ ~a|
|R Rl+1
Substitute to obtain;
l l
 
V = 41 2q
+ q
P al P (x) + q P (1) a P (x)
l l
R Rl+1 Rl+a
Here x = cos()

9
 2q
V = 4
P al P (x)
l+1 l
l=2; even R

The lowest non-zero term in the series represents the multipole moment of the charge distri-
bution.

9 Example2
In the development of scattering problems, we will need the expansion of a plane wave in
spherical coordinates. Note that the [lane wave is symmetric in azimuthal angle. Thus;
~
eikz = eik~r = cn Yn0
P
n

Use ortho-normality to project out the coefficients

R
cn = 2 sin() d Yn0 eikr cos()
0

Apply the integral representation of the spherical Bessel function;

R
jn (z) = (1/2)(i)n 0
eiz cos() Pn () sin() d

This gives;

eikr cos() = il [4(2l + 1)]1/2 jl (kr) Yl0 ()

P
l

By the addition theorem;

eikr cos() = 4
P m (
jl (kr) Ylm (k)Y r)
l
lm

Figure 5 shows the geometry of the addition theorem.

10 Angular momentum
~ The angular momentum
Neglecting mass, the momentum operator in QM has the form i~.
operator then has the form;

~ = ~r i~
L ~ ~r ~p

The energy operator in spherical coordinates is;

10
 z k

Figure 5: The geometry of the addition theorem

~2 2 = ~2 [2 + 2 ]
T = 2M 2M r

where the Laplacian operator is divided into operations on the radial variable, r, and the
angular variables, (, ). In Cartesian coordinates;

Lx = , i~[y z ]
z y
Lx = , i~[x x ]
z z
Lx = , i~[x y ]
y z
The map between Cartesian and spherical coordinates is;

x = r sin() cos()

y = r sin() sin()

z = r cos()

Make a change of variables for the Cartesian angular momentum operators above.

Lx = i~[sin() + cot() cos() ]

Ly = i~[cos() + cot() sin() ]

11
 Lz = i~

~ L
Then look at L2 = L ~

L2 = ~2 [ 1 [sin() ] + 1 2 ]
sin() sin () 2
2

This can be obtained by using;

~ = i~ ~r
L ~

~
~ = r ~r ~r2

r r
Then we have the eigenvalue equation;

L2 Ylm (, ) = ~2 l(l + 1)Ylm (, )

The commutation rules for the angular momentum operator can be worked out as well using
the above description of the operators.

[Li , Lj ] = iijk ~ Lk

L = Lx iLy

[Lz , L ] = ~L

[L+ , L ] = 2Lz

Lz Ylm = ~ m Ylm

Lz L Ylm = ~(m 1)L Ylm

L2 commutes with L so from the above, L+ (L ) is a raising (lowering) operator on the

eigenvalue m in the eigenfunction Ylm

11 Vector spherical harmonics

When we deal with the vector Laplacian, it is useful to introduce vector spherical harmonics
of the form;

~m
l =
p 1 ~ m
LY l
l(l + 1)

12
The operator L will be the same as the angular momentum operator without the multiplying
~ is defined by L
constant of ~. Thus L ~ = i[~r ].
~ Then;

L2 = [ 1 (sin() ) + 1 2 ]
sin() sin () 2
2

L2 Ylm = l(l + 1)Ylm

All other operations follow in the same way as was the case of the angular momentum terms.
The ~ functions as defined above are ortho-normal.

N2 = 1 R
~ m ) LY
d(LY ~ m
l(l + 1) l l

Then;

~ m = i~r
LY ~ Ym
l l

Substitute into the integrand to obtain;

N2 = 1 R
~ m Y
d r 2 [Y ~ m ]
l(l + 1) l l

Integrate by parts to produce a 2 operator operating on Ylm . The result is;

d ~m

~m
R
l l = ll m,m

In the case of non-Cartesian coordinates the vector Laplacian should be used when operating
on a vector field. We will then use the expansion of a vector field in a new set of functions
called vector spherical harmonics. Define 3 vector operators to span 3-D space;

~ m = r[( l + 1 )1/2 Y m ] +[ 1 Ylm im

V ] + [ Y m]
l 2l + 1 l
[(l + 1)(2l + 1)]1/2 [(l + 1)(2l + 1)]1/2 sin() l
~ m = r[( l )1/2 Y m ] + [ 1 Ylm im
W 1/2 ] + [ Y m]
l 2l + 1 l
[(l)(2l + 1)] [(l)(2l + 1)]1/2 sin() l
m Ylm i Ylm
~m
= [ ] + [ ]
l
[(l)(2l + 1)]1/2 sin() [(l)(2l + 1)]1/2
These functions satisfy orthogonality and are normalized;

~m B
~ m = AB ll mm
R
d Al l

~,B
Where A ~ are any of the above functions.

13
12 Expansion of a vector field
Any vector field can be expanded as;

F~ =
~ 1 + L
~ 2 + curlL/i
~ 3

To see this note that any vector can be written;

~ 1 +
F = ~ G
~

~ 3
Then let Q = i~r2 iL ~ 4 . Then take
~ Q.
~

~ = i[~r ]
Use the definition L ~ to show that;

~ Q ~ L
~ 2 +
~ = L ~
i 3

In spherical coordinates;

Qr = ir2 4
r
Q = 1 3 (1/r) 4
sin()
Q = 3 1 4
r sin()
One can then demonstrate that the above equations may be solved for 2 , 3 , 4 . Then for
the vector potential, A, we can write;

~ = ~ 2 +
~ 1 + L ~
~ L
A i 3
However, when considering the magnetic field using Coulomb gauge, ~ A~ = 0. Given the
~ the only zero component is represented by,
definition of the operator, L, ~ [L
~ 2] = 0

13 Example3
Apply the vector spherical harmonics to the calculation of the magnetic field of a current
loop. From Amperes law in steady state;

~ B
~ =
~
~ A
~ = J~

14
 z

V = F( )
a
V=0
x

Figure 6: The potential inside a hemisphere over a flat plane

~ is the magnetic field, A is the vector potential, and J~ is the current density.
In the above, B
We choose the Coulomb gauge, ~ A ~ = 0. This reduces the above to;

~ R(r)~
A ml

~ separates to obtain;
Then 2 A

1 [r 2 d2 R + (2r) dR ] = 1m [L2 ]m = l(l + 1)

R dr 2 dr l l

~ or by using
The remainder of the solution is obtained by matching boundary conditions for A
the Greens function. When m = 0 (azimuthal symmetry) and for the solution to remain
finite as r the solution will have the form;

i dYl0
F~l = al (1/r l+1) [
P
]
(l(l + 1))1/2 d

14 Example4
Consider the potential due to a semi-circular screw head on a flat conducting plane. The
geometry is illustrated in Figure 6. We look for solutions to Laplaces equation, 2 V = 0,
for the potential in a spherical coordinate system. In general the solution must have the form;

Alm r l Ylm +
r (l+1)
P P
V = lm Blm
lm

Note that while the general solution should include the Legendre functions of the 2nd kind
as well as functions of the 1st kind the boundary condition that the solution be finite along
the z axis excludes the Ql functions. The solution is also symmetric in azimuthal angle so
m = 0. If we wish solutions inside the a hollow screw head, then the solution must be finite
at r = 0. In this case the solution takes the form;

Al r l Pl
P
V =
l

15
For the potential to vanish when = /2 choose only values of l = 1, 3, 5, .

An r 2n+1 P2n+1 (cos())

P
V =
n=0

Finally, match the boundary condition, F (x) with x = cos() when r = a, by projecting out
of the sum the coefficients, An
1
An = 2n(2n+1)
+ 1 R dxF (x) P
2n+1 (x)
2a 1

On the other hand suppose we seek a solution for r > a. In this case, the solution should
have the form;

Bl r (l+1) Pl (x)
P
V =

As above, we wish the solution finite on the z axis, azimuthally symmetric and finite as
r . We will need to change the boundary conditions slightly to find a stable solution in
spherical coordinates as the application of Cauchy (Neumann) conditions must be applied on
a closed surface. The boundary problem is really due to a mixture of boundary conditions
on spherical and cylindrical surfaces. Thus choose to put the potential on the surface at
z = 0 to a constant value which we can choose as V = 0. There will be another potential,
V = V0 on a plane which we take as r . Then the boundary conditions are ;

V = 0 for z = 0, and r = a

V = Ez z as r

The boundary condition V = Ez z is really a Neumann condition specifying the electric

~ = V
field, E ~ . Then we must have a solution of the form;

Bn r (2n+2) P2n+1 (x)

P
V = Ez [r cos()] +
n

15 Example5
As another example, consider a dielectric sphere of dielectric constant, , in a uniform field
~ = Ez z. The boundary conditions are;
in the z direction, E

V = Ez z = Ez r cos() as r

~ is continuous at r = a
Tangential E

16
 V = E_{z} z V = E_{z} z
z

a
x

V = E_{z} z V = E_{z} z

Figure 7: The potential of a dielectric sphere in a uniform electric field

~ = E
Perpendicular D ~ is continuous at r = a

The boundary conditions at r = a are derived for a dielectric without surface charge or
currents from Gauss and Amperes laws.

The solution must have the form after applying azimuthal symmetry and requiring a finite
solution on the z axis (x = cos());

Al r l Pl (x) + Bl r (l+1) Pl (x)

P P
V =
l l

To match the boundary condition as r require the solution to be;

Bl r (l+1) Pl (x)
P
V = Ez r cos() +
l
~
Then match the boundary conditions at r = a. The boundary condition that tangential E
is continuous is the same as requiring a continuous potential at r = a.

Al al Pl = Ez r P1 + Bl a(l+1) Pl
P P
l

For Dr to be continuous;

V |in = V |out
r r
Al al1 Pl = Bl (l + 1) a(l+2) Pl = Ez P1
P P
r

Solve for the coefficients;

A0 = 0

B0 = 0

A1 = +3 E
r 2 z

17
 (r 1)a3
B1 = r + 2 Ez
All other coefficients vanish. The potential is;

Inside r < a

V = 3Ez r cos()
r +2

Outside r > a

V = Ez r cos() + [ r + 1 ] E a3 cos()
r + 2 z r2

16 Multipole EM fields
We look for solutions to Maxwells equations with the time component suppressed. Let
k = /c. The Helmholtz equation is;

~ + k2 B
2 B ~ = 0

The operator, 2 in the above is the vector Laplacian. We want a set of spherical harmonic
~ ( or for E.
solutions (multipoles) for B ~ Each component of B satisfies the Helmholtz equa-
~ = 0 and we want to find solutions in terms of the spherical Hankel
tion as well as div B
functions as these have the correct boundary conditions as r . The solution is then
expected to have the form;

~ ~ Ym
P
B alm hl L l

In the above hl is the spherical Hankel function giving the radial dependence as r and
the angular term is the vector spherical harmonic ~ml whose divergence vanishes. This solu-
tion then satisfies the Helmholtz equation has appropriate boundary conditions as r ,
~ B
and has ~ = 0. The solution for the electric field is similar, and to get a fully complete set
of functions we need to combine the fields, (ie the magnetic field as above and the magnetic
~ E
field component due to the electric field ~ = i B, ~ and of course the electric component
due to the magnetic field). Therefore the complete solution takes the form;

~ = ~m ~ ~m
P
B A(E)lm fl l (i/k) A(M)lm gl l
lm
~ = ~ fl
~m ~m
P
E (i/k)A(E)lm l + A(M)lm gl l
lm

18
For a wave moving radially outward, fl , gl are the Hankel functions of the first kind. The
coefficient A(E) [A(M)] represents the amount of electric [magnetic] multipole in the solution.
A(E) means the source is due to a charge density and A(M) is due to a magnetic moment
density.

19