Separation of variables in spherical coordinates

October 30, 2015

We will make important use of the separation of variables in spherical coordinates, because the separation
ends up giving us a series in terms of powers of r. For an isolated system, this means the expansion gives a
way to approximate the field far from the source.

1 Separation of variables in spherical coordinates

In spherical coordinates the Laplace equation takes the form

∂2V
   
1 ∂ 2 ∂V 1 ∂ ∂V 1
r + sin θ + =0
r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ ∂ϕ2
We apply the same steps as we did for Cartesian coordinates, but the difference in the form of the Laplace
equation raises some new issues.

1.1 Step 1: Substitute the separation assumption

We assume a solution of the form
V (r, θ, ϕ) = R (r) Θ (θ) Φ (ϕ)
and substitute.

1.2 Step 2: Find the separated equations

Dividing by RΘΦ, we have

1 d2 Φ
   
1 d 2 dR 1 1 d dΘ 1
r + 2 sin θ + 2 =0
r2 R dr dr r sin θ Θ dθ dθ r2 sin θ Φ dϕ2

This time, nothing separates until we multiply by r2 sin2 θ,

r2 sin2 θ d 1 d2 Φ
   
2 dR sin θ d dΘ
2
r + sin θ + =0
r R dr dr Θ dθ dθ Φ dϕ2
and even now only the Φ term has completely separated.
Take the partial derivative with respect to ϕ. Since the first two terms depend only on r and θ, this
shows that
1 d2 Φ
 
∂
= 0
∂ϕ Φ dϕ2
and therefore
1 d2 Φ
= −m2
Φ dϕ2

1
Here we usually choose the constant to be negative because solutions involving the full range of ϕ from 0 to
2π must be periodic. Lesser ranges may have either sign. The Φ equation is therefore

d2 Φ
+ m2 Φ = 0
dϕ2
The remainder of the Laplace equation is now

sin2 θ d
   
2 dR sin θ d dΘ
r + sin θ − m2 = 0
R dr dr Θ dθ dθ

so dividing by sin2 θ separates the remaining two variables,

m2
   
1 d 2 dR 1 1 d dΘ
r + sin θ − =0
R dr dr Θ sin θ dθ dθ sin2 θ
Differentiating with respect to r and θ respectively show that
 
1 d 2 dR
r = a
R dr dr
m2
 
1 1 d dΘ
sin θ − = −a
Θ sin θ dθ dθ sin2 θ
where the two constants must add to zero.

1.3 Step 3: Solving the equations

The simplest equation is the one for Φ, with solution

Φ = C sin mϕ + D cos mϕ

or more commonly
Φ (ϕ) = Am eimϕ
for positive and negative m.
Rewriting the R equation as  
d dR
r2 − aR = 0
dr dr
Notice that rescaling r leaves this equation unchanged. This is a clue that powers of r may work. Substituting
a power-law solution, R (r) = rl ,
l

!
d 2d r
r − arl = 0
dr dr
d l+1

l r − arl = 0
dr
(l (l + 1) − a) rl = 0

and with a = l (l + 1) we have a solution for every number l. Notice that there are two values of l that give
the same value for asince the quadratic equation,

l2 + l − a = 0

2
has two solutions. Let l have some value k, so that a = k (k + 1). Then the value l = − (k + 1) gives the
same value, a = (− (k + 1)) (− (k + 1) + 1) = k (k + 1).
Setting a = l (l + 1), we have only the θ equation remaining,

m2
   
1 d dΘ
sin θ + l (l + 1) − Θ = 0 (1)
sin θ dθ dθ sin2 θ
The solutions to the θ equation with m = 0 are called the Legendre polynomials. Solutions with general m
may be found by differentiating the Legendre polynomials, giving the associated Legendre polynomials. We
consider only the m = 0 case. To start, we define a new variable,

x = cos θ

Then
d dx d
=
dθ dθ dx
d
= − sin θ
dx
and therefore
d 1 d
= −
dx sin θ dθ
d
sin2 θ dΘ


Thus, with m = 0, and replacing the derivatives, eq.(1) becomes dx dx + l (l + 1) Θ = 0. Replacing
sin2 θ = 1 − cos2 θ = 1 − x2 , we have the Legendre equation,
 
d
dΘ
1 − x2 + l (l + 1) Θ = 0 (2)
dx dx
The solutions are polynomials, Pl (x) = Pl (cos θ).
For example, suppose Θ takes the form

P2 (x) = a0 + a1 x

Then substituting,
 
d 2 dP2 (x)


0 = 1−x + l (l + 1) P2 (x)
dx dx
 
d 2 d (a0 + a1 x)


= 1−x + l (l + 1) (a0 + a1 x)
dx dx
d
a1 1 − x2 + l (l + 1) (a0 + a1 x)



=
dx
= −2a1 x + l (l + 1) a0 + l (l + 1) a1 x
= l (l + 1) a0 + (l (l + 1) a1 − 2a1 ) x

This gives two equations,

l (l + 1) a0 = 0
l (l + 1) a1 − 2a1 = 0

The first gives either a0 = 0 or l (l + 1) = 0, while the second gives either a1 = 0 or l (l + 1) = 2.

This equation for l has two possible solutions, l = 1 or l = −2. Since we cannot have both a0 and a1
vanishing, we get two possibilities:
P0 = a0 , l = −1 or 0

3
or
P1 = a1 x, l = −2 or 1
The remaining constants are chosen so that Pl (1) = 1. The presence of two solutions for l for the same
Legendre polynomial means that for each Pl there will be two powers of r. The solutions will be of the form
   
1 1
V (r, θ) = r0 + P0 (cos θ) + r + 2 P1 (cos θ) + · · ·
r r

where

P0 (cos θ) = 1
P1 (cos θ) = cos θ

Like the harmonic functions, there are many relationships among the Legendre polynomials. The most
important for us now is the orthogonality relationship,
π
ˆ2
2
Pl (cos θ) Pl0 (cos θ) sin θdθ = δll0
2l + 1
0

2
Notice that the right side may be written as either 2l+1 δll0 or as 2l02+1 δll0 since it is only nonzero when l = l0 .
We may also write this in terms of x = cos θ. With dx = − sin θdθ, we have

ˆ1
2
Pl (x) Pl0 (x) dx = δll0
2l + 1
−1

When m is nonzero the solutions are called associated Legendre polynomials, Plm (x). The range of m is
from −l to +l.

1.4 Step 4: Put it all together

For general m, the full solution is
∞ X
l  
X Blm
V (r, θ, ϕ) = Alm rl + l+1 Plm (cos θ) eimϕ
r
l=0 m=−l

We will only consider m = 0 cases, which apply to problems with azimuthal symmetry. Setting Al0 = Al
and Blm = Bl ,
∞  
X Bl
V (r, θ) = Al rl + l+1 Pl (cos θ) (3)
r
l=0

The constants Alm and Blm are uniquely determined by the boundary conditions.

2 Fitting boundary conditions in spherical coordinates

2.1 Example: Piecewise constant potential on hemispheres
Let the region of interest be the interior of a sphere of radius R. Let the potential be V0 on the upper
hemisphere, and −V0 on the lower hemisphere,
 π   π 
V (R) = V0 Θ −θ −Θ θ−
2 2

4
We require the potential to be nonsingular everywhere within the sphere.
To find the potential, we may immediately write
∞  
X
l Bl
V (r, θ) = Al r + l+1 Pl (cos θ)
r
l=0

and impose the boundary conditions. First, notice that finiteness everywhere means that we cannot have
the rB
l+1 terms because they diverge at r = 0. Therefore, we set Bl = 0 for all l. The remaining condition at
l

r = 0 is
 π ∞
  π  X
V0 Θ −θ −Θ θ− = Al Rl Pl (cos θ)
2 2
l=0
We multiply this by Pk (cos θ) and integrate from 0 to π,
ˆπ  π ∞ ˆπ
  π  X
l
V0 Θ −θ −Θ θ− Pk (cos θ) sin θdθ = Al R Pl (cos θ) Pk (cos θ) sin θdθ
2 2
0 l=0 0
π
ˆ2 ˆπ ∞
X 2Al Rl
V0 Pk (cos θ) sin θdθ − V0 Pk (cos θ) sin θdθ = δkl
2l + 1
0 π l=0
2

ˆ1 ˆ0
2Ak Rk
V0 Pk (x) dx − V0 Pk (x) dx =
2k + 1
0 −1

First, for k = 0, P0 (x) = 1 and the left side vanishes and A0 = 0.

For the k > 0 integrals on the left we use the symmetry of the Legendre polynomials. Change variable
from x to −x in the second integral to write
ˆ ˆ0 ˆ ˆ0
 1   1 

V0  Pk (x) dx − Pk (x) dx = V0  Pk (x) dx + Pk (−x) dx

0 −1 0 1
ˆ1 ˆ1
 

= V0  Pk (x) dx − Pk (−x) dx

0 0
ˆ1
= V0 (Pk (x) dx − Pk (−x)) dx
0

We know that the odd Legendre polynomials are polynomials in odd powers of x, and the even are even.
Therefore,
Pk (−x) = Pk (x) k even
Pk (−x) = −Pk (x) k odd
and the difference we need vanishes for all even polynomials. For the odd cases, the left side becomes
ˆ1 ˆ1
V0 (Pk (x) dx − Pk (−x)) dx = 2V0 Pk (x) dxdx
0 0

Now, among many identities for the Legendre polynomials, we find

ˆ
Pk+1 (x) − Pk−1 (x)
Pk (x) dx =
2k + 1

5
so the integral we need becomes

ˆ1 1
Pk+1 (x) − Pk−1 (x) 
2V0 Pk (x) dx = 2V0
2k + 1 
0
0
2V0
= [(Pk+1 (1) − Pk−1 (1)) − (Pk+1 (0) − Pk−1 (0))]
2k + 1
2V0
= − (Pk+1 (0) − Pk−1 (0))
2k + 1
since Pk (1) = 1 for all k. This still requires some work to determine the value of Legendre polynomials at
x = 0. For now we just leave the answer in terms of these values, so for all odd k,

2V0 2Ak Rk
− (Pk+1 (0) − Pk−1 (0)) =
2k + 1 2k + 1
V0
Ak = (Pk−1 (0) − Pk+1 (0))
Rk
The potential at all points inside the sphere is therefore,
∞
X rl
V (r, θ) = V0 (Pl−1 (0) − Pl+1 (0)) Pl (cos θ)
Rl
l=0

2.2 Example: Varying potential on a sphere

Let a sphere of radius R have potential

V (r = R, θ, ϕ) = V0 cos2 θ

Find the potential everywhere inside and outisde the sphere.

For the interior solution, we again must have the potential finite at r = 0, so we immediately set Bl = 0
and write the potential as
∞
X
V (r, θ) = Al rl Pl (cos θ)
l=0

To fit the boundary condition at r = R, we require

∞
X
V0 cos2 θ = Al Rl Pl (cos θ)
l=0

or, with x = cos θ,

∞
X
V 0 x2 = Al Rl Pl (x)
l=0

Because the left hand side is a simple, low order polynomial, it is easiest to write the left side in Legendre
polynomials. We only need polynomials up to order x2 and only even ones. The two relevant Legendre
polynomials are therefore

P0 (x) = 1
1
3x2 − 1


P2 (x) =
2

6
Solving for x2 ,
2 1
x2 = P2 (x) +
3 3
2 1
= P2 (x) + P0 (x)
3 3
The boundary condition therefore requires
∞
2V0 V0 X
P2 (x) + P0 (x) = Al Rl Pl (x)
3 3
l=0

and orthogonality of the polynomials implies, matching terms,

V0
A0 =
3
2V0
A2 R 2 =
3
The potential everywhere inside the sphere is
V0 0 2V0 2
V (r, θ) = r P0 (cos θ) + r P2 (cos θ)
3 3R2
V0 V0 2r2 1
3 cos2 θ − 1


= +
3  3 R2 2
r2

V0 2


= 1 + 2 3 cos θ − 1
3 R
The potential outside the sphere has the same boundary condition at r = R, but requires Al = 0.
Therefore, the boundary condition at R will be
∞
2V0 V0 X Bl
P2 (x) + P0 (x) = Pl (x)
3 3 Rl+1
l=0

so that
B0 V0
=
R 3
B2 2V0
=
R3 3
and the potential becomes
∞
X Bl
V (r, θ) = Pl (x)
rl+1
l=0
B0 B2
= P0 (x) + 3 P2 (x)
r  r
R2

V0 R
1 + 2 3 cos2 θ − 1


=
3r r
Notice that the interior and exterior solutions agree at r = R.

2.3 Example: Varying potential on a sphere (a more challenging example)

Let a sphere of radius R have potential
V (r = R, θ, ϕ) = V0 sin4 θ
Find the potential everywhere inside and outisde the sphere.

7
Inside:
Since we have spherical boundary conditions, it is easiest to use the spherical separation, and since the
problem has azimuthal symmetry, we may use the solution to the Laplace equation in the form
∞  
X
l Bl
V (r, θ) = Al r + l+1 Pl (cos θ)
r
l=0

Our boundary conditions are

V (0) = f inite
V (R, θ, ϕ) = V0 sin4 θ

The first of these conditions shows that Bl = 0 for all l. The outer boundary condition becomes
∞
X
V0 sin4 θ = Al Rl Pl (x)
l=0

Essentially, we must express V0 sin4 θ in terms of Legendre polynomials.

Let x = cos θ. Then noting that we may write

2
V0 sin4 θ = V0 1 − cos2 θ

2
= V 0 1 − x2
= V0 1 − 2x2 + x4

we should be able to write the potential on the sphere in terms of Legendre polynomials of even order less
than or equal to x4 . Using higher order polynomials would introduce undesired higher powers of x. The
three relevant polynomials are:

P0 (x) = 1
1
3x2 − 1


P2 (x) =
2
1
35x4 − 30x2 + 3


P4 (x) =
8
Rearranging and combining to produce x4 − 2x2 + 1, we start with
1
x4 8P4 (x) + 30x2 − 3


=
35
1
x2 = (2P2 (x) + 1)
3
1
x4 − 2x2 + 1 8P4 (x) + 30x2 − 3 − 2x2 + 1


=
35    
8 30 3
= P4 (x) + − 2 x2 + 1 −
35 35 35
 
8 6 14 32
= P4 (x) + − x2 +
35 7 7 35
 
8 8 1 32
= P4 (x) − (2P2 (x) + 1) +
35 7 3 35
8 16 8 32
= P4 (x) − P2 (x) − +
35 21 21 35

8
  
8 16 8 4 1
= P4 (x) − P2 (x) + −
35 21 7 5 3
 
8 16 8 7
= P4 (x) − P2 (x) +
35 21 7 15
8 16 8
= P4 (x) − P2 (x) + P0 (x)
35 21 15
Therefore, we must solve
  X∞
8 16 8
V0 P4 (x) − P2 (x) + P0 (x) = Al Rl Pl (x)
35 21 15
l=0

for the coefficients Al . To do this, we use the orthogonality relation,

ˆ 1
2
Pl (x) Pl0 (x) dx = δll0
−1 2l +1
Multiply our equation by an arbitrary Pk (x) for any fixed k, and integrate over all x,
ˆ 1  ∞ ˆ 1
8 16 8 X
V0 P4 (x) − P2 (x) + P0 (x) Pk (x) dx = Al R l Pl (x) Pk (x) dx
−1 35 21 15 −1
l=0
ˆ ˆ ˆ ∞
8V0 1 16V0 1 8V0 1 X 2
P4 (x) Pk (x) dx − P2 (x) Pk (x) dx + P0 (x) Pk (x) dx = Al R l δlk
35 −1 21 −1 15 −1 2l + 1
l=0
8V0 2 16V0 2 8V0 2 2
δk4 − δk2 + δ0k = Ak R k
35 2k + 1 21 2k + 1 15 1 2k + 1
8V0 2 16V0 2 8V0 2 2
δk4 − δk2 + δ0k = Ak R k
35 9 21 5 15 1 2k + 1
The left side vanishes unless k = 0, 2 or 4, and in these cases we have
8V0
A0 =
15
16V0
A2 = −
21R2
8V0
A4 =
35R4
The full solution everywhere inside the sphere is found by putting these coefficients back into the general
form, giving the final potential as
∞  
X
l Bl
V (r, θ) = Al r + l+1 Pl (cos θ)
r
l=0
= A0 r0 P0 (cos θ) + A2 r2 P2 (cos θ) + A4 r4 P4 (cos θ)
8V0 0 16V0 2 8V0 4
= r P0 (cos θ) − 2
r P2 (cos θ) + r P4 (cos θ)
15  21R 35R4 
1 2r2 r4
= 8V0 − 2
P2 (cos θ) + P4 (cos θ)
15 21R 35R4
16r2 8r4
 
8
= V0 − P 2 (cos θ) + P 4 (cos θ)
15 21R2 35R4
We immediately verify that if we set r = R and use the form of sin4 θ in terms of Legendre polynomials, we
recover the boundary condition. At the center of the sphere, we find (setting r = 0) that the potential is
8
V (0) = 15 V0 .

9
Aside: Verify Theorem 3.1.4 If we integrate the potential over the surface of the sphere (using Wolfram
integrator),

ˆπ ˆ2π ˆπ
4
sin θdθ dϕV0 sin θ = 2πV0 sin5 θdθ
0 0 0
 π
5 5 1
cos3 θ −

= 2πV0− cos θ + cos 5θ 
8 48 80 0
   
5 5 1 5 5 1
= 2πV0 − + − − + −
8 48 80 8 48 80
 
5 5 1 5 5 1
= 2πV0 − + + − +
8 48 80 8 48 80
 
5 5 1
= 2πV0 − +
4 24 40
 
150 25 3
= 2πV0 − +
120 120 120
 
128
= 2πV0
120
32π
= V0
15
so that the potential at the center of the sphere is

ˆπ ˆ2π
8 1
Vcenter = V0 = V (R, θ, ϕ) R2 sin θdθdϕ
15 4πR2
0 0

as required by the theorem of section 3.1.4 in Griffiths.

Outside:
The exterior solution is similar, but this time the boundary conditions are

V (∞) = 0
V (R, θ, ϕ) = V0 sin4 θ

which means that we must have Al = 0 for all l. The second condition now reads
  X ∞
8 16 8 Bl
V0 P4 (x) − P2 (x) + P0 (x) = Pl (x)
35 21 15 Rl+1
l=0

Bl
so we need only replace Al Rl of the interior solution by Rl+1
. This gives

2R3 R5
 
R
V (r, θ) = 8V0 − 3
P2 (cos θ) + P4 (cos θ)
15r 21r 35r5

2.4 Example: Extending the solution for a disk

Here is a somewhat different example of the use of the series solution and uniqueness. Although this is
not a boundary condition in the usual sense, it is still enough information to find all of the coefficients

10
and construct the potential everywhere. We have found previously that the potential on the z-axis above a
circular disk of radius R lying in the xy-plane is
σ p 2 
V (z) = z + R2 − z
20
Since we have axial symmetry, the full solution
∞  
X
l Bl
V (r, θ) = Al r + l+1 Pl (cos θ)
r
l=0

must agree with this when we set z = r and θ = 0:

 ∞  
σ p 2 X Bl
z +R −z 2  = Al r l + Pl (cos θ)|θ=0
20 z=r rl+1
l=0
∞  
σ  p  X Bl
r 2 + R2 − r = Al r l +
20 rl+1
l=0

where we have used Pl (cos 0) = Pl (1) = 1, for all l.

The problem is simplest for the region where r > R. In this case, we may expand the square root in a
Taylor series as r
p
2 2
R2
r +R =r 1+ 2
r
1/2
We need the Taylor series for (1 + x) for small x. Looking at the first few derivatives
1/2
f (x) =
(1 + x)
1 −1/2
f (1) (x) = (1 + x)
2
 
1 1 −3/2
f (2) (x) = − (1 + x)
2 2
  
3 1 1 −5/2
f (3) (x) = − − (1 + x)
2 2 2

This is enough to see the pattern. After k differentiations, we will have

k
(−1) −(2k−1)/2
f (k) (x) = − (2k − 3) (2k − 5) · · · 1 (1 + x)
2k
k
(−1) (2k − 3)! −(2k−1)/2
= − k (1 + x)
2 (2k − 2) (2k − 4) (2k − 6) · · · 2
k
(−1) (2k − 3)! −(2k−1)/2
= − k k−1
(1 + x)
2 2 (k − 1) (k − 2) (k − 3) · · · 1
k
(−1) (2k − 3)! −(2k−1)/2
= − (1 + x)
2k 2k−1 (k − 1)!

so the full Taylor series is

∞
X (−1) (2k − 3)!k
1/2 1
(1 + x) = 1+ x− xk
2 22k−1 (k − 1)!
k=2

11
 R2
Setting x = r2 ,the right side of our equality therefore becomes
∞ ∞ 
! !
k
1 R2 X (−1) (2k − 3)! R2k

σ X
l Bl
r 1+ − −r = Al r + l+1
20 2 r2 22k−1 (k − 1)! r2k r
k=2 l=0
∞ ∞
!
k
σ R2 X (−1) (2k − 3)! R2k
 
X Bl
− = Al rl + l+1
0 4r 22k (k − 1)! r2k−1 r
k=2 l=0

This means that Al = 0 for all l, and for the Bl we match term by term to find:

σR2
B0 =
40
∞ k ∞
σ X (−1) (2k − 3)! R2k X Bl
− =
0 22k (k − 1)! r2k−1 rl+1
k=2 l=1

Note that we have only odd powers of r on the left, so only even l will occur on the right. Let l = 2k − 2 on
the right. Then for all k > 1
∞ k ∞
σ X (−1) (2k − 3)! R2k X B2k−2
− =
0 22k (k − 1)! r2k−1 r2k−1
k=2 k=2
k
σ (−1) (2k − 3)! 2k
− R = B2k−2
0 22k (k − 1)!

All odd Bl vanish. The potential everywhere for r > R (including off axis!) is therefore
∞ k 2k−1
σR2

σR X (−1) (2k − 3)! R
V (r, θ) = − P2k (cos θ)
40 r 0 22k (k − 1)! r
k=2

The solution for r < R may be found in a similar way, although it requires two subcases.

3 Exercises
3.1 Separation of variables
Separate the Laplace equation in cylindrical coordinates to find the differential equations for three functions.
Solve the equations for the ϕ and z directions. The radial equation gives Bessel functions.

3.2 Legendre polynomials

Suppose Θ (θ) takes the form
Pl (x) = a0 + a1 x + a2 x2 + a3 x3
Substitute into the Legendre equation, eq.(2) and find all allowed combinations of solutions for a1 , a2 , a3 and
l. This will lead you to P0 , P1 , P2 and P3 . Choose the normalizations so that Pl (1) = 1.

3.3 Boundary conditions on a sphere

Consider a sphere of radius R held at a potential

V (R, θ) = V0 cos θ sin2 θ

Find the potential everywhere, both inside and outside the sphere.

12