Lecture 6 Notes, Electromagnetic Theory I: 1. Associated Legendre Polynomials
Lecture 6 Notes, Electromagnetic Theory I: 1. Associated Legendre Polynomials
Lecture 6 Notes, Electromagnetic Theory I: 1. Associated Legendre Polynomials
][
d P ml x
d
m2
1 x 2
l l1
P lm x =0 where x=cos
2
dx
dx
1x
- Once this equation is solved, the general solution for the Laplace equation in spherical
coordinates will have the form:
l
r , ,= Al r B l r
l1
m=0
Am 0 Bm0 P l
cos
Al r l B l r l 1 Am e i m B m ei m P ml cos
m0,l
- Encouraged by the Rodrigues' form of the solution to the m = 0 Legendre equation, we try a
d
solution of the form P ml ( x)=(1 x 2)m / 2 dx g ( x) and substitute in:
m
0=m x 1x 2 m /2
m
d m 1
dm
2 d
2
2 m /21
g
m
g
x
1x
m
g x1 x 2m / 2
m 1
m
m
dx
dx
dx
m1
d m2
2 m/ 2 d
g
m2
x
1x
g x
dx m2
dx m1
dm
dm
[ l l1 ] 1x 2 m /2 m g x m2 1x 2 m/ 21 m g x
dx
dx
1x 2 m /21
0=1x
d m d2
d m1
dm
g
x
2
[
m1
]
x
g
x
[
l
l1mm1
]
gx
dx m dx 2
dx m1
dx m
- We want to move as much as possible inside the derivatives. We can do this by using the
product rule.
m
dm
m!
d m k
u
v=
k ! mk ! dx mk u
dx m
k=0
][ ]
- Let us try to apply this to rearrange the first term in the differential equation.
dk
v
dx k
- If u=1 x 2 then
that:
d
dx
u=2 x ,
d2
dx2
[ ]
dm
dm
d m 1
d m2
2
2
1x
v
=1x
v
2
m
x
v
mm1
v
dx m
dx m
dx m 1
dx m2
[ ]
m
m1
m 2
d
d
d
d
v = m 1x 2 v2 m x
v m m1
v
m
m1
m 2
dx
dx
dx
dx
d
- Now if we set v = dx g ( x) this product rule expansion becomes:
2
2
dm d2
dm
d m1
dm
2 d
1x
g x = m 1x 2 g x 2 m x
g x mm1
g x
dx m dx 2
dx
dx
dx m1
dx m
2
- We can use this product rule to rearrange the first term of the Legendre equation so it
becomes:
0=
2
dm
d m1
dm
dm
2 d
1
x
x
2
x
g
xl
l
1
g
x2
m
g x
dx m
dx 2
dx m1
dx m
dx m
- Now we want to use the product rule expansion on the second term of the Legendre equation.
du
d
With u=x , dx =1 , all higher derivatives are zero and v = dx g ( x) , the general product rule
becomes:
] [
] [
] [
dm
d
d m1
dm
x
g
=
x
g
m
g x
dx m dx
dx m1
dx m
- After rearranging:
x
] [
d m1
dm
d
dm
g
=
x
g
m
g x
m1
m
m
dx
dx
dx
dx
- Substitute in:
m
[
[ (
0=
d
d
d
1 x 2 2 g x2 x
g x l l1 g x
m
dx
dx
dx
0=
dm d
d
(1 x 2) g ( x) +l (l+1) g ( x )
m
dx
dx
dx
- It is worth noting that if this is true for m = 0, then it will automatically be true for all m. So
we can set m = 0:
0=
d
d
1 x 2 g x l l1 g x
dx
dx
- Now this is just the m = 0 Legendre equation, which we have already solved. We found the
solutions to be the ordinary Legendre polynomials, Pl(x), so that g x=P l x .
- We now have the solution to the full Legendre equation:
P ml x=1x 2 m /2
dm
Pl x
dx m
- There is an arbitrary phase factor that we have assumed to be one for simplicity, but is
conventionally set to 1m , so that the associated Legendre functions become:
P ml ( x)=(1)m (1x 2 )m /2
d
Pl (x )
m
dx
m+l
(1)m
2 m /2 d
(
1x
)
( x 21)l
l
m+l
2 l!
dx
dm d
d
1x 2 g x l l1 g x
m
dx
dx dx
that the highest power of x inside the brackets is l, so that m cannot be greater than l, otherwise
the equation would be trivially satisfied and thus there would be no unique solution. In
summary, both l and m are integers and the possible m are -l, -(l-1), ..., 0, ..., (l-1), l.
- Several useful mathematical relations involving the associated Legendre functions can be
found (the derivations are left to the interested student).
m
Pm
l x=1
lm! m
P x
lm! l
lm1 P ml1 x =2 l1 x P ml x lm P ml 1 x
x=lm x P ml xl m Plm1 x
1 x 2 P m1
l
2
x 1
d m
m
m
Pl x =l x P l xl m P l 1 x
dx
P ml ' x P ml x dx= 2 l1
lm! l ' l
lm!
2. Spherical Harmonics
- With the full Legendre equation now solved, the general solution of the Laplace equation in
spherical coordinates has been found:
r , ,= Al r l B l r l 1 Am 0B m0 P m=0
cos
l
l =0
l
Al r B l r
l =0 m=1
l 1
i m
Am e
i m
B m e
Pl cos
r , ,= Al r l Bl rl 1 Am e i m Bm ei m P ml cos
l =0 m=0
- An alternate way of presenting this is to let m sum from -l to l and thus combine the Am and Bm
terms with the other constants:
r , ,=
Al , m r l B l ,m r l1 e i m P ml cos
l =0 m=l
- The normalizing terms are included in the undetermined constants Alm and Blm. When a
boundary condition is applied, these constants become specified and the normalizing terms
result naturally. However, to make the math cleaner, we can explicitly pull the normalizing
terms out of the constants in advance.
r , ,=
Al , m r l B l ,m r l1
l =0 m=l
2 l1 lm ! i m m
e Pl cos
4 lm !
- It is now evident that Plm(cos ) times its normalization term forms a complete set of
orthonormal functions in the variable , indexed by l for a fixed m, and eim times its
normalization term forms a complete set of orthonormal functions in the variable , indexed by
m. The product of both thus creates orthonormal functions that spans all angles of the unit
circle. These are known as spherical harmonics Ylm:
Y lm ( , )=
2 l +1 (lm)! i m m
e P l (cos ) so that
4 (l+m)!
(r , , )=
( Al , m r l +Bl , m r l 1)Y lm ( , )
l=0 m =l
where x=cos
0 1
or
- The validity of this equation can be checked trivially by expressing the spherical harmonics in
terms of its definition and applying the orthonormality of the associated Legendre functions and
the complex exponentials.
- Using the definitions of the spherical harmonics, the patient student can work out the explicit
analytic form for any given l and m. The lowest-order spherical harmonics are especially
simple and are typically tabulated in textbooks.
- Because the spherical harmonics form an orthonormal set of equations, an arbitrary function
f , can be expanded in terms of spherical harmonics:
f ( , )=
l =1 m=l
- There are several useful special cases for spherical harmonics that we should keep in mind.
- If m = 0, the spherical harmonic does not depend on the azimuthal angle and the associated
Legendre function reduces down to a Legendre polynomial:
Y l , 0 ( , )=
2 l+1
Pl (cos )
4
- If m = l, the spherical harmonic's dependence on polar angle becomes especially simple, just
the sine function to the mth power, so there is only one maximum in this direction:
Y lm (0, )=
0 if m0
2 l+1
if m=0
4
and Y lm ( , )=
0 if m0
2 l+1
(1)l if m=0
4
(r , , )=
l=0 m =l
- The region of interest includes the origin, thus Blm = 0 to keep the solution from blowing up
there:
( r , , )=
l=0 m =l
Alm r l Y lm ( , )
V ( , )=
Alm r l0 Y lm ( , )
l=0 m =l
A r = V ( , ) Y *lm ( , )sin d d
l
lm 0
0 0
( r , , )=
l=0 m =l
Alm r Y lm ( , ) where
l
0
Alm=r
V ( , )Y *lm( , )sin d d
0
(r , , )=
l=0 m =l
Alm
()
r
Y lm ( , ) where
r0
P l cos =
4
Y * ' ,'Y lm ,
2 l1 m=l l m
r , ,=
1
rl
r0
1
= l1 P l cos if r r 0 and
= l 1
P l cos if r r 0
rr 0 l =0 r 0
rr0 l=0 r
- The equations still hold when the point r0 is off the z-axis if the Legendre polynomials become
functions of the angle between the observation point r and the source point r0.
r l0
1
r
1
=
P cos if r r 0 and
=
P cos if r r 0
rr0 l=0 r l01 l
rr0 l=0 r l 1 l
- This equation is not terribly useful because the angle depends on the spherical coordinates.
- We can use the addition theorem to expand the ordinary Legendre polynomial so that the
expression is a function of spherical coordinates rather than a function of the angle between
coordinate vectors:
l
1
1
rl *
=4
Y l m ( ' , ') Y lm ( , ) if r <r 0 and
l+1
rr0
l =0 m=l 2 l+1 r 0
l
r l0
1
1
=4 l+1
Y *l m (' , ') Y lm( , ) if r >r 0
rr
2
l+1
0
l =0 m=l r
- This will be useful if we want to know the potential produced by multiple point charges in
explicit coordinates.
5. The Laplace Equation in Cylindrical Coordinates
- Earlier, we solved the Laplace equation in cylindrical coordinates for the special case of when
the boundary conditions are uniform in the z-dimension, and the problem reduced to polar
coordinates.
- We now revisit the problem for when the boundary conditions are not uniform in the zdimension.
- The Laplace equation in cylindrical coordinates is:
2 =0
1 2 2
=0
2 z 2
1
R
1 2 Q
2 Z z
R Z z 2
RQ
=0
2
z2
- Divide by R Q Z z :
1 1
R
1 1 2 Q
1 2 Z z
=0
R
Q 2 2
Z z z2
- The last term depends only on z and the other terms do not so to be valid for all z, they must be
related by a constant:
1 1
R
1 1 2 Q 2
1 2 Z z
2
k =0 where k =
R
Q 2 2
Z z z2
- Multiply the left equation by 2 :
R
1 2 Q 2 2
1 2 Z z
2
=0
k
=
where
R
Q 2
Z z z2
- The second term depends now only on and the rest of the terms do not, so we set it to a
constant. All the partial derivatives become regular derivatives as the functions are of one
variable only now:
d R
1 d 2 Q
1 d2Z z
d
2
2
2
2 2
k =0 where =
, k =
Q d 2
Z z d z2
R d
d
1 d
d R
2
k 2 2 R=0 where
d
d
d 2 Q
d2 Zz
2
=
Q
=k 2 Z z
,
2
2
d
dz
- In order to get the most general solution, we must cover all four cases:
- Case 1: If =0 and k =0 then the solutions are:
R = A0,0B 0,0 ln , Q=C 0,0 D0,0 and
Z z = F 0, 0G 0, 0 z
, Q=C ,0 e i D , 0 ei and
R = A ,0 B ,0
Z z = F , 0G ,0 z
Q=C , k e
kz
D ,k e
k z
Z z = F , k e G , k e
- The differential equation involving R must be solved by trying a series solution. Once found,
the general solution for this case will have the form:
, , z = R , k C , k e
D , k e
kz
k z
F , k e G , k e
0 k0
1 d
d R x
2
x
1 2 R x =0
xdx
dx
x
j =0
j=0
j=0
a j x j2 j2 2 a j x j =0
- Remove the first two terms of the sum on the left and combine the remaining sums:
a 0 x 2 2 2a 1 x 1 12 2 x j a j2 j22 2 a j =0
j=0
- Every power of x is independent, thus the coefficient of every power must vanish:
a 0 2 2 =0 , a 1 12 2 =0 , a j2 =
1
aj
j22 2
1
a
j2 j22 j
- Because all odd powers are zero, we can iterate over all integers and rewrite the indices:
a 2 j =
1
a 2 j 1 for j =1, 2, 3,...
4 j j
R , k = J k where J x= a 2 j 1 x
j=1
convention, a 0=
2 j1
, a 2 j =
1
a
and by
4 j j 2 j 1
1
in terms of the Gamma function.
2 1
- One might think that the other solution to the R equation is J-, but it turns out that this is not
linearly dependent from J when is an integer and we must find another solution.
- Traditionally, the Neumann function, or Bessel function of the second kind is taken as the
other linearly independent solution. It is defined by:
N x =
J x cos J x
sin x
- Useful properties, including the recurrence relations, can be easily derived or found in
textbooks.
- The general solution for R, in terms of the Bessel functions, becomes:
R , k = A , k J k B ,k N k
- The general solution to the Laplace equation in cylindrical coordinates for case 4 becomes:
, , z = A , k J k B , k N k C , k e i D , k ei F , k e k z G , k ek z
0 k0
- The most general solution to the Laplace equation in cylindrical coordinates is the sum of all
the solutions of all the possible cases:
( , , z )=( A0,0+B 0,0 ln)(C 0,0+ D0,0 )( F 0, 0+G0, 0 z )
+ ( A , 0 + B ,0 )(C ,0 e i +D ,0 ei )(F ,0+G ,0 z )
0
k z
k0
+ ( A , k J (k )+B ,k N ( k ))( C ,k e i + D , k ei )( F ,k e k z +G , k ek z )
0 k0
- To be useful, we must find orthogonal sets of functions involving the Bessel functions so that
we can use the orthogonality condition to invert equations involving the boundary conditions.
The functions:
J xn
can be shown to be orthogonal on the interval (0, a) for fixed and n = 1, 2, 3,... where xn are
the roots that satisfy J x n =0 .
- The orthogonality condition for these functions can be shown to be:
a
J
0
x n'
a2
J x n d = [ J 1 x n ] 2 n' n
a
a
2
f = A n J x n
a
j=1
where
A n=
2
2
a J
2
1
x n
f J
x a
n
i
i
A , 0 B , 0 C ,0 e D , 0 e
F , 0G ,0 z
0
A , k J k B ,k N k C ,k e
kz
kz
k z
k0
0 k0
D ,k e
k z
F , k e G , k e
- The region where we need a valid solution includes the origin, so several of the coefficients
must be zero to keep the solution from blowing up. This leads to:
, , z =C 0,0D 0,0 F 0,0 G0,0 z
C ,0 e i D , 0 e i F ,0 G , 0 z
0
J k C ,k e i D ,k ei F , k e k zG ,k ek z
0 k0
- The region where we need a valid solution spans the full circle of angles for , so we require
the function to be single-valued:
, , z = , 2 , z
- This equality must hold for all values of all of the independent variables, thus each term must
match. This requirement leads to: D 0,0 =0 , D0,k =0 , and =m where m=0,1,2,3,. ..
, , z = F 0,0 G0,0 z
m C m ,0 e i m D m , 0 ei m F m , 0G m , 0 z
m0
J 0 k F 0,k e k z G0, k ek z
k0
J m k C m ,k e i m Dm ,k e i m F m , k e k z Gm ,k ek z
m0 k 0
J 0 k F 0,k G0, k
k0
J m k C m ,k e i m D m , k ei m F m , k Gm , k
m0 k 0
J 0 k F 0, k sinh k z
k0
J m k C m ,k e i m Dm ,k e i m sinh k z
m0 k 0
- All the m = 0 terms now have a form for which they can be combined with the m 0 terms:
m
, , z = C m ,0 e
im
i m
D m , 0 e
J m k C m, k e
m k0
im
i m
D m , k e
sinh k z
- The first term on the right is just a special case of the last term when k = 0 (close to zero, Jm
has an asymptotic form that is proportional to m). They can be combined:
, , z = J m k C m , k ei m D m , k ei m sinh k z
m
im
Dm ,k e
i m
sinh k z
- This can only hold true for all values of the independent variable and z if the coefficients are
zero:
x mn
a
a C
, , z = J m x mn
m=0 n=1
m ,n
e i m Dm ,n ei m sinh x mn
z
a
a C
V , = J m x mn
m=0 n=1
m, n
ei m D m , n ei m sinh x mn
L
a
- This is a Fourier series in and a Fourier-Bessel series in . We find the coefficients in the
usual way. Multiply both sides by an exponential and integrate:
2
V ,e
i m '
d = J m
m=0 n=1
x mn
a
x mn
L
a
V ,ei m d =2 J m
n=1
x a C
mn
m ,n
sinh x mn
L
a
J m
0
x mn '
a
V , e
i m
n=1 0
J m x mn '
0
a
n=1
i m
L
J m x mn d C m , n sinh x mn
a
a
a
a2
J x n d = [ J 1 x n ]2 n ' n
a
a
2
V , ei m d d =2
V ,e
J m x mn
0
d d =2 J m x mn'
a
a
L
[ J x ]2 n ' n C m, n sinh x mn
2 m1 m n
a
d d = a 2 [ J m1 x m n ]2 C m ,n sinh x mn
L
a
L
2
a
a
C m ,n =
d
d V , J m x mn ei m
2
2
a
a [ J m1 x mn ] 0
0
csch x mn
- In the exact same process the other set of coefficients is found to be:
L
2
a
a
D m , n=
d
d V , J m x mn ei m
2
2
a
a [ J m1 x m n] 0
0
csch x mn
( , , z )= J m x mn
m=0 n=1
where
( La )
csch x mn
C m ,n =
( )
z
(C m , n ei m +C *m , n ei m )sinh x mn
a
a
a [ J m+1( x m n
d d V ( , ) J m x mn ei m
( a)
)]
2