Chapter 7

Laplaces Equation
Last update: 17 Dec 2009.
Syllabus section;
7. Laplaces equation. Uniqueness under suitable boundary conditions. Separation of variables. Twodimensional solutions in Cartesian and polar coordinates. Axisymmetric spherical harmonic solutions.

7.1 The Laplace and Poisson equations

Let (r) be a scalar field in three dimensions, as in previous chapters. Laplaces equation is simply
2 = 0

(7.1)

where, as we met in Chapter 3.6, 2 () div(grad ); 2 is called the Laplacian operator, or

just the Laplacian.
Remember from before, if is a scalar field, is a vector field, and then taking div of that gives us
another scalar field, so Laplaces equation is a scalar equation.
In Cartesian x, y, z coordinates, things are simple: we recall the definitions from Chapter 3,
=

i+
j+
k
x
y
z

and
F =

F1 F2 F3
+
+
x
y
z

Putting the first into the second above, Laplaces equation becomes
2

2 2 2
+ 2 + 2 =0
x2
y
z

(7.2)

Note that if we are using other coordinates (e.g. cylindrical polars or spherical polars) we must use results for
grad and div in those coordinates from Chapter 5, so it will look different; we look at those later.
Laplaces equation often occurs as follows: suppose we have a conservative vector field F, so that F =
for some scalar field as in Chapter 4.7; then if F = 0 this gives Laplaces equation 2 = 0.
89

Aside: Laplaces equation is the simplest and most basic example of one of the three types of secondorder linear partial differential equations (PDEs), known as the elliptic type. Laplaces equation is a linear
homogeneous equation.
A generalisation of Laplaces equation is Poissons equation which is
2 = f (r)
, where f (r) is a given scalar field. Laplaces equation is clearly a special case of Poissons where f = 0 at
all points in the volume of interest.
The basic examples of the other types are the wave equation
1 2 f
= 2 f ,
c2 t 2
where c is constant (usually the speed of sound or light) and the heat equation or diffusion equation

f
= 2 f
t
where f is temperature in a solid, and is constant. We met the heat equation with a single spatial variable
in Example 6.8 on Fourier series.
In maths, the wave equation is an example of the hyperbolic PDE and and the heat equation is the
parabolic PDE. These names are potentially confusing since the solutions have nothing to do with ellipses,
parabolas, or hyperbolas, but this is just a shorthand because the powers and signs in the equations look
somewhat similar to the equations for ellipsoids, paraboloids and hyperboloids from Chapter 1.
Going back to our gravitational and electromagnetic examples of conservative fields, and using the Divergence Theorem
Z
Z
V

F dV =

F.dS

to obtain Gausss results, we see that if F = 0 everywhere there are no sources inside the volume, which
for gravity means that there is no mass there and for electric field means that there is no (net) charge. Hence
Laplaces equation describes the gravitational potential in regions where there is no matter, and the electric
potential in regions where there are no charges.
If instead there is a net charge density , the electric field satisfies
E =

1
(r)
0

where 0 is a constant of nature. (This is one of the four Maxwells equations). Combining this with E =
gives
1
2 = (r).
0
This is an example of Poissons equation as we met above. Laplaces equation is of course a special case of
Poissons equation, in which the function on the right-hand side is zero throughout the volume of interest.
Laplaces and Poissons equations are very important, both because of their occurrence in many physics
applications, and because they are the basic examples of elliptic PDEs. We are now going to spend the rest
of this chapter considering some solutions of Laplaces equation in 2 dimensions.
We can see directly that there are some simple solutions of Laplaces equation, e.g.

constant

= c
90

= x

= y
= xy

= x 2 y2
etc

These clearly are solutions, by direct evaluation of 2 from Eq. 7.2. There are in fact an infinite number of
general solutions to Laplaces equation, which are known as harmonic functions.
As is common in differential equations, to find the solution in a specific case we need to be given some
boundary conditions. Next, we will prove that under suitable boundary conditions the solution of Poissons
(or Laplaces) equation is unique. We shall then investigate what the solutions actually are in some simple
cases, in each of Cartesian, cylindrical and spherical polar coordinates.

7.2

Uniqueness of Solutions to Poissons (and Laplaces) Equation

Note that 2 is a linear operator: that is

2 ( 1 + 2 ) = 2 1 + 2 2

for any two scalar fields 1 , 2 and any two constants , . Hence if 1 and 2 are both solutions of
Laplaces equation, so is 1 + 2 . Also, if is a solution of Poissons equation and is a solution of
Laplaces equation, + is also a solution of Poissons equation, for the same f (r).
Theorem 7.1 Suppose that 2 u = f (r) throughout some closed volume V , f (r) being some specified function
of r, and that the value of u is specified at every point on the surface S bounding volume V . Then, if a solution
u(r) exists to this problem, it is unique.
Proof
Suppose that u1 and u2 are two scalar fields which both solve the above problem. Define w u 1 u2 . Then,
2 w = 0
w=0
on S.

in V,
since u1 = u2 on S

Now consider the volume integral

Z

|w|2 dV

=
=

(ww) w2 w dV

(ww) dV 0

w(w).dS

(w).(w) dV

ZV

(using Eq. 3.4

since 2 w = 0 everywherein V

(by the Divergence Theorem)

= 0

because w = 0 everywhere on S. Now, the integrand on the LHS is a square therefore is always non-negative,
and its integral is zero. This can only happen if w is zero throughout V (otherwise, if w was non-zero
anywhere in V , the whole integral would be positive because there cannot be any negative bits in the integral
to cancel the positive part, i.e. a contradiction).
91

Now w = 0 throughout V means w is constant throughout V . But w = 0 on the boundary, so w = 0

throughout V . Hence u1 = u2 throughout V. So the solution is unique. Q.E.D.
Note that we have proved uniqueness for Poissons equation, and Laplaces is a special case.
Aside: It is fairly clear that the final step in the displayed calculation also works if, instead of w = 0 on
the boundary, w.n = 0 where n is the normal to the surface S. This corresponds to being given a boundary
value for u.n on the boundary, instead of the value of u itself. Moreover, it still works if at each point either
u or u.n, is specified. The case where u is given are called Dirichlet boundary conditions, and the case
where u.n is given is called Neumann boundary conditions. If we only have Neumann conditions, our
w above is still a constant but not necessarily zero, so the solution u is only unique up to addition of any
arbitrary constant. We will only deal with Dirichlet boundary conditions from here on, but you may meet the
Neumann conditions in later courses.
The virtue of this theorem is that it gives us a licence to make whatever assumptions or guesses we like,
provided we can justify them afterwards by showing both Laplaces (or Poissons) equation and the boundary
conditions are satisfied: if they are, the solution we found must be the right one, even if our method was not
very rigorous.
Having proved uniqueness, we now demonstrate how to actually find solutions of Laplaces equation in
some simple situations. In general (r) can depend on all three coordinates, but we will confine ourselves to
cases depending on two of the three coordinates: we will study the three most common coordinate systems
as before:
In Cartesian coordinates, we will take (x, y), so does not depend on z.

In cylindrical polar coordinates, we will take U( , ) so U does not depend on z again, and we relabel
to U to avoid confusion with the angle .

In spherical polar coordinates, we will take U(r, ), so U does not depend on and we have rotational
symmetry around the z axis.
The first two of these cases provide us with a nice simple interpretation. For (x, y) or ( , ) the
obvious choices of boundary conditions are on walls at fixed x, y. But we can then choose to forget the z
direction and think of a planar problem. Now imagine as a varying height h. Then solving Laplaces
equation in 2D subject to boundary conditions is like holding a rubber sheet up with its edges stuck to an
arbitrarily shaped warped rigid hoop. The hoop fixes the height at the boundary, while the rubber tries to
minimize the total area.
For (r, ) though, we still need to think in three dimensions, and the simple interpretation no longer
applies.
Note: A physical example in three dimensions is as follows: suppose we take a uniform solid object
(of arbitrary shape), and attach a large number of tiny thermostat-controlled heater/coolers to the surface, and
turn their thermostats to some smooth varying function on the surface. The temperature inside, T (r), will
obey the heat equation
T
= 2 T ,
t
with a constant and boundary conditions set by our thermostats. If we wait a long enough time so the
temperature distribution inside reaches a steady state, the LHS above will then be zero, so then the temperature
inside will solve Laplaces equation with the given surface settings as the boundary condition.
The choice of coordinates will be adapted to the geometry of the domain of interest and its boundaries,
which usually makes calculations easier. For rectangular boundaries we use Cartesians, for circles in the plane
92

or circular cylinders in three dimensions we use plane or cylindrical polars, and for spherical boundaries we
use spherical polars. For example, one may need to calculate the electrostatic potential outside a charged
sphere. This would be very messy in Cartesian coordinates, and is much simpler if we use spherical polar
coordinates instead. (This was one of the main reasons for studying Chapter 5 )

7.3

2-D solutions of Laplaces equation in Cartesian coordinates

We first develop a general method for finding solutions = (x, y) to Laplaces equation in a rectangular
domain, with given boundary conditions on all four edges of the rectangle. In Cartesian coordinates, as we
saw above, Laplaces equation is
2 = () =

2 2 2
+ 2 + 2 =0 .
x2
y
z

(7.3)

We will now try looking for a solution of the form

(x, y) = X(x)Y (y).
where X(x) is some function of x only, and Y (y) is some function of y only. Such a solution is called a
separable solution. We cannot justify this in advance, but if it works then the uniqueness theorem tells us we
are OK. It is possible to prove that any solution can be written as a sum (possibly an infinite sum) of separable
solutions, but this is beyond the scope of this course.
Substituting the above into (7.3) gives
d 2Y
d2 X
Y + X 2 = 0.
2
dx
dy
Dividing this by XY gives

1 d 2Y
1 d2 X
=

.
X dx2
Y dy2

Now, the left-hand side is a function of x only, and the right-hand side is a function of y only. This can
only be satisfied if both sides are an unknown constant (call that constant , real, with the minus sign for
convenience).
Note: to prove the constant, the above equation is true at any x, y: so consider the above equation along a
line (x0 , y) with fixed x = x0 and varying y. The LHS is fixed, so the RHS must therefore be independent of
y, i.e. constant. The same argument with y0 fixed shows the LHS is constant.
Thus we have

d2 X
+ X = 0 and
dx2

d 2Y
Y = 0.
dy2

If k 6= 0, these equations are the differential equations for trigonometric and hyperbolic functions, which we
met in chapter 1, so we know their general solutions as follows:
If is positive, define k so that = k2 for convenience, and the solution is
X = A cos kx + B sinkx,

Y = C cosh ky + D sinhky,

where A, B,C, D are any constants. Multiplying these together,

= (A cos kx + B sinkx) (C cosh ky + D sinhky) .
93

(7.4)

If is negative, define k =

and then the solution is

X = A cosh kx + Bsinhkx,

Y = C cos ky + Dsin
ky.

B,
D are different constants. Then
C,
where A,

= A coshkx + Bsinhkx
C cos ky + Dsin
ky .

(7.5)

Note: in each of these solutions there is usually one more constant than we really need. For example if
in (7.4) AC 6= 0 we can write
= AC (cos kx + B/A sinkx) (coshky + D/C sinhky)

using just three constants AC, B/A and D/C: this means that in examples, one of the four constants can
usually be set to 1. One way to do this is to write (7.4) as
= L sin (kx + M) sinh (ky + N)
for some constants L, M, and N. Usually this works fine, except it does not cover the case where D = 0.
Finally, we need to deal separately with the case with = 0:
we can integrate each equation twice to arrive at
= (A0 x + B0)(C0 y + D0) ,

(7.6)

with more constants A0 , B0 , C0 and D0 . It is often convenient to multiply this out and re-write it as
= + x + y + xy
with , , , as alternative constants.
Remember, from linearity, any sum of any of the above functions with any k and any constants is also a
solution of Laplaces equation. So, if we are given a boundary condition, and we can pick-and-mix any sum
of the above functions which does satisfy all the given boundary conditions, then we have solved the problem
(and our solution is unique). If we are lucky, a particular one of the separable solutions will do this, as we see
in the next example.
Example 7.1. Find the solution of
2

2 2
+ 2 =0
x2
y

()

in the domain D: 0 x a, 0 y b, given boundary conditions = 0 on x = 0, on y = 0 and on x = a,

and = sin(p x/a) on y = b, for some integer p.
We note here that is zero along three of the sides, and non-zero along the top side with y = b. Also
since sin 0 = 0 and sin(p a/a) = 0, is zero at the points (0, b) and (a, b) so the given function is continuous
at the corners.
Can we satisfy the boundary conditions in this case with one of the separable solutions? We consider
them one by one. Clearly (7.6) will not work since it doesnt contain a sin. The form (7.4) is more promising,
since if we take that equation and choose
A=0

B=1

k=

p
a

in that, the first bracket becomes sin(p x/a) which is the function we want on the boundary y = b. Now we
just need to choose C, D to make the second bracket in 7.4 equal zero at y = 0 and 1 at y = b; this gives us
C cosh0 + D sinh0 = 0
94

C cosh(n b/a) + D sinh(n b/a) = 1

and the first of these implies C = 0, then the second gives D = 1/ sinh(n b/a).
So finally,
(x, y) = sin

p x
p y
p b
sinh
/ sinh
a
a
a

This satisfies all the boundary conditions and Laplaces equation, so it is the unique solution.
In the above Example, we chose a sin in the boundary condition to make it easy: but for more general
boundary conditions, using just one separable solution will not work.
However, since Laplaces equation is linear, we can add together separable solutions to get a more general
solution. In many cases, including the Cartesian one, it is possible to prove that every solution can be written
as a sum of separable solutions (this is called completeness of the separable solutions).
In the Cartesian case we would need to introduce different values of A for each k etc., which we typically
would denote Ak . Since k can take any value, the sum of separable solutions can in general become an
integral1 over k, but for the rectangular boundaries in the example above we will only need to take integer
values of p, call it n, so the general solution becomes
= (A0 x + B0 )(C0 y + D0)

+ (An cosn x/a + Bn sin n x/a)(Cn coshn y/a + Dn sinh n y/a)

n=1

+ (an cosh n x/b + bn sinhn x/b)(cn cos n y/b + dn sin n y/b)

n=1

We note that the sin n x/a terms vanish at x = 0 and x = a so they will fit Dirichlet boundary conditions
which are zero on those boundaries. If multiplied by a sinhn y they also vanish on y = 0 so are non-zero only
on y = b: to get similar forms which are zero at y = b and non-zero at y = 0 we need to take a combination
of sinh n y and coshn y which is zero at y = b, which will turn out to look like sinh n (b y)/b by the
addition formula.
( The cos n x/a terms are not zero on the boundary, but have vanishing derivative n. = / x at
x = 0 and x = a, so they will fit Neumann boundary conditions which are zero on those boundaries. Since we
will stick to Dirichlet problems as examples in this course, we will find we are using only the sine terms ).
Similar remarks apply to the other two sides at x = 0 and x = a, just with x y and a b.
From these remarks, we can see that in order to fit boundary conditions we now typically have to work
out the Fourier series for the functions given on the boundaries. Again, since solutions of Laplaces equation
can be added together, it is perfectly valid to work out solutions that have = 0 on three of the four sides of
the rectangle, and fit the (non-zero) conditions on the fourth boundary, do this for each choice of fourth side
in turn, and add the four resulting solutions together. That is why in examples, we usually take cases where
= 0 on three of the four sides of a rectangle.
Moreover, in practice conditions like = 0 can be applied to each separable solution in turn, rather than
considering the sum as a whole: this may look unjustified, but uniqueness allows us to justify it if the final
answer works. Using these ideas let us look again at the generalized version of the previous example.
1 This

leads to the use of Fourier transforms, which is the next step, beyond this course, in Fourier methods

95

Example 7.2. Consider the previous example but with = g(x) on y = b.

On y = b, we try a linear combination of solutions of the form found above (keeping the conditions
derived from the other parts of the boundary):
(x, y) =

Dn sinh

n=1

n x
n y
sin
.
a
a

This is automatically a solution of Laplaces equation.

However, putting in y = b along the boundary gives us
(x, b) =

Dn sinh

n=1

n b
n x
sin
.
a
a

and this is a function of x only, so we can write this as

(x, b) = En sin

n x
a

()

with En Dn sinh(n b/a).

Finding the coefficients En in equation () is a standard problem in (arbitrary range) Fourier series [Comment: now you see why we did section 7.5]. Multiplying both sides by sin(m x/a) and using the result
that

Z a
m x
n x
a/2 if m = n
sin
sin
dx =
0 if m 6= n
a
a
0
gives

m x
2 a
dx.
g(x) sin
a 0
a
Evaluating this integral for all m and putting it back gives us a solution to the original problem,
Em =

En

sinh(n b/a) sinh (n y/a)sin

n=1

n x
.
a

By uniqueness, we have found the solution.

We still have a couple more issues to deal with. So far, we have seen how to solve the problem as a
Fourier series when the boundary conditions are zero on three sides and non-zero on any one side.
If the boundary conditions are non-zero on all four sides but still zero at all four corners, we can solve
this just by breaking it into four problems (each non-zero on one side) giving four solutions 1 , 2 , 3 , 4 ,
and adding the four solutions, by linearity of Laplaces equation.
If the four corners are all one constant value, just subtract this constant from the boundary conditions,
solve, and add it back to the final solution.
Finally, we have to deal with the case where the given boundary conditions are different (but still continuous) at the four corners. This can be dealt with by Eq. 7.6 above: it is straightforward to choose our
four constants , , , to give a solution (call it 0 ) which matches the given boundary values at all four
corners, by starting with the (0, 0) corner then the (0, 1), etc. Next, we subtract that 0 (x, y) from all the
given boundary conditions to get a new set of boundary conditions for 1 + 2 + 3 + 4 ; solve 1 to 4 by
treating the four sides separately as above; and finally add all five solutions 0 + . . . + 4 to get the answer.
96

0
2y

sin y

sin y

sin x

0
sin x

Figure 7.1: Left: boundary conditions on (x, y). Right: boundary conditions after subtracting off 0 = xy.

Example 7.3. Consider a rectangle with 0 x 2, 0 y 1, and boundary values for (x, 0) = sin x
etc. as shown at the left diagram in Figure 7.1.
First we solve for the coefficients in 0 (x, y) = + x + y + xy (which is just another way to write
(7.6)) so as to fit the corners.
0 (0, 0) = 0 = 0,
0 (2, 0) = 0 = 0,
0 (0, 1) = 0 = 0,
0 (2, 1) = 2 = 1, so
0 = xy
Subtracting this off leaves the boundary conditions at the right of Figure 7.1. We can now get (x, 0) correct
along the bottom with
sinh( (1 y)) sin x
1 (x, y) =
sinh( )
(this is like example 7.1) and get (0, y) right with
2 (x, y) =
The full solution is

sinh( (2 x)) sin y

.
sinh(2 )

= 0 + 1 + 2 .

We should also note that there are 2-dimensional harmonics which are not of the above form, but are
mixed power series in x and y. These are less useful for general problems with rectangular boundaries, but
may be useful in other contexts. They will of course be expressible as infinite series of the form above, but
not as finite sums of such terms. One way to obtain them is to rewrite the z-independent cylindrical polar
harmonics (see below) in Cartesians, using the relation of Cartesians and plane polars.
Exercise 7.1. Find (x, y) in 0 < x < , 0 < y < 1, satisfying the following conditions:
2 = 0 in 0 < x < , 0 < y < 1,
= sin x on y = 0
and = 0 on the other three sides of the rectangle. Is the solution unique?

7.4

2-D solutions of Laplaces equation in cylindrical polar coordinates

We now look at cylindrical polars; this is the natural choice where the boundary conditions are given on a
circle or cylinder. It will turn out a bit simpler than Cartesians since there are no corners to worry about on
the boundary.
97

We also change our label for our scalar field from to U, to avoid confusion with angle (of course, this
is just a re-labelling and makes no real difference).
From chapter 5, in cylindrical polar coordinates, the grad of a scalar field U is
U =

U
U
1 U
e +
e +
ez

and the divergence of F = F e + F e + Fz ez is



1 ( F ) F ( Fz )
F =
.
+
+

z
Putting these together we obtain







1 U
1
U
U

U div(U) =
+
+
,

z
z
2

which simplifies to
2U =

1 2U 2U
+ 2 .
2 2
z

Consider the case when everything in the problem is independent of z, so U = U( , ). Once again we
seek a separable solution, this time we will write it as
U( , ) = R( )S( ) .
where R and S are functions to be found. Putting this into 2 and rearranging gives


d
dR
1 d2 S

=
R d
d
S d 2
Once again, the LHS is a function of only and the RHS is a function of only , so by the same argument as
before, both sides are some (unknown) constant, call it again.
The differential equation for S is then

d2 S
+ S = 0,
d 2

(7.7)

which is just the now familiar equation encountered in the Cartesian case. If > 0, it has the general solution

S( ) = A cos( ) + B sin( ) .
If < 0 we would similarly have
S( ) = A cosh(

) + B sinh(

) .

But, the solution must be the same if we change by 2 , since the two values and + 2 represent
the same point in space; which means that
the sinh and cosh terms with < 0 are forbidden. The sin
and cos terms will obey this condition iff is an integer. Thus, the only allowed values are = m2
where m is a positive integer (without loss of generality) and we can write the solution for a particular m as
S = Am cos m + Bm sin m .
Now, putting = m2 back into the R differential equation gives


d
dR

= m2 R.
d
d
98

We guess a power-law solution R = C q for some constant q; substituting gives

q2 = m 2
so q = m. This is two independent solutions for q, and each has its own constant, so we write
R( ) = Cm m + Dm m
, and again Cm , Dm are constants; multiplying out the S and R, we have a solution for U of the form


U( , ) = (Am cos m + Bm sin m ) Cm m + Dm m .
for some integer m > 0.

The case = 0 is a special case: then we integrate twice so R = C0 ln + D0 , and S = A0 + B0 . As in

the Cartesian case each of these solutions has more constants than we need in any given example. In most
cases we set A0 = 0 by uniqueness on adding 2 to ; ( but note there are special cases where it is acceptable
for U not to be unique, provided U is unique. This happens in fluid dynamics, for example, where we are
interested in the fluid velocity u = U rather than the potential U itself. In this case we require that U be
single valued, which allows us also to use the A0 term).
So, the general solution of Laplaces equation in cylindrical polars is a linear combination of all these
above for the m = 0 case and every integer m > 0,
U( , ) = (A0 + B0)(C0 ln + D0 ) +

(Am cos m + Bm sin m )

m=1

Cm m + Dm m .

(7.8)

Note that this form implies that boundary conditions on a surface of fixed lead to a Fourier series problem
in (once the terms in A0 have been found). However, in many cases we need only a finite number of terms
and can use intelligent guesswork (essentially, including only terms with the same values of m which appear
in the boundary conditions) to choose a suitable form for U.
Also, note the presence of both positive and negative powers of : if we are solving a problem inside a
circle with boundary condition given on the circle, we will require all D m to be zero for m 1 so the solution
is sensible at = 0. Alternatively, we can be given boundary conditions on a circle and find a solution outside
the circle, requiring the solution to be well-behaved at large ; then all Cm are zero for m 1. If we solve in
an annulus in between two circles of given radii, with boundary conditions given on both the inner and outer
circles, then we will need both Cm and Dm terms to match the given functions on both boundaries.
Example 7.4. Consider 2U( , ) = 0 with the boundary conditions U(1, ) = 2 sin2 on the unit
circle, and U ln at large .
First look at the general solution 7.8. That does not contain a sin 2 , but since 2 sin2 = 1 cos2 , the
latter form does look like a sum of two terms in 7.8: a constant (m = 0) terms and a cos 2 term which
looks like an m = 2 term; so we can (correctly) guess that the same is true of the solution, i.e. we choose all
Am . . . Dm coefficients with m = 1 and m 3 to be zero, so the infinite sum becomes just one term with m = 2.
We also set B2 = 0 since our boundary condition only has cos 2 not sin 2 .
The large- condition implies also A0 = 0, and also C2 = 0 since we dont want a +2 term at large .
Writing out 7.8 without all those zeros leaves us with our educated guess solution as
U = B0 D0 + B0C0 ln + A2 D2 cos(2 ) 2
. This has several redundant constants, and we can just rewrite it as
U = + ln + cos(2 ) 2
99

Finally, matching the given function on the circle = 1 gives us = 1, = 1, and the large condition
gives us = 1, so the unique solution is
U( , ) = 1 + ln

cos 2
.
2

We can check this simply: it is a particular case of 7.8 so it does satisfy Laplaces equation. And it matches
the given boundary conditions on = 1 and large ; so it is the unique solution.
Exercise 7.2. Consider the region D defined by a b, 0 , < z < . Sketch the region in
a plane perpendicular to the z-axis which lies in D. On the boundaries = a, = 0 and = , U = 0 while
on the boundary = b, U = sin . Find the solution U of Laplaces equation in D, independent of z, which
satisfies these boundary conditions.
[You may assume that on 0

sin =

16k

(4k2 1)2 sin 2k .]

k=1

2
Note: Finally, it is also worth noting that solutions like n cos(n ) can also be expanded as polynomials
in x, y: for example, cos(4 ) = 8 cos4 8 cos2 + 1, and 4 = (x2 + y2 )2 , therefore a bit of arithmetic leads
to 4 cos 4 x4 6x2 y2 + y4 , and you can use the Cartesian formula to check that 2 of that is zero. These
may occasionally be useful, but they rapidly get unmanageable for large n.

7.5

Axisymmetric solutions of Laplaces equation in spherical polar

coordinates

Now we consider what to do in problems with a naturally spherical geometry. First, we need to work out
what 2U is in spherical polar coordinates.
As before, we have

2U div(U).

which is true in any coordinate system. Now in spherical polar coordinates,

U =

1 U
U
1 U
e +
e
er +
r
r
r sin

and the divergence of F = Fr er + F e + F e is



1
(r2 sin Fr ) (r sin F ) (rF )
+
F = 2
+
.
r sin
r

Putting these together we obtain

 





1

1 U
2U = 2
r2 sin
+
sin
+
,
r sin r
r

sin
which simplifies to
 




1
1
U
1 2U
2 U
U= 2
r
+
sin
+ 2
.
r r
r
sin

sin 2
2

100

Many problems are axisymmetric that is, there is no dependence on the coordinate. In such cases
U = U(r, ) and (anything)/ = 0. As in the previous cases, we proceed by seeking a separable solution:
U(r, ) = R(r)S( ).
[different meanings from the R and S in the last section]. Thus 2U = 0 becomes


 
 
1
S
1
2 R
r
S
+
sin
R =0

r2 r
r
sin

which rearranges to





1
R

1
S
r2
=
sin
.
R(r) r
r
S( ) sin

Once again, the left-hand side is a function of r only, and the right-hand side is a function of only. But they
are equal, and so they must both be some constant, say . Thus


d
2 dR
r
R = 0
(7.9)
dr
dr
and

1 d
sin d

sin

dS
d

+ S = 0.

(7.10)

We consider equation (7.10) first. If we define w = cos , then

d
1 d
=
,
dw sin d
so equation (7.10) can be written in the form


d
2 dS
(1 w )
+ S = 0.
dw
dw
which is called Legendres differential equation. We see in the next Section that only Legendre polynomial
solutions P` (w) = P` (cos ) are allowed, i.e. the cases where = `(` + 1) and ` is an integer, and P` is the
Legendre polynomial of order `.
Going back to equation (7.9), inserting = `(` + 1) then R(r) satisfies


d
2 dR
r
`(` + 1)R = 0.
dr
dr

(7.11)

We try looking for a power-law solution, R = Ar p of this: we find

p(p + 1)Ar p = `(` + 1)Ar p
i.e. p(p + 1) = `(` + 1). Given `, this is a quadratic equation for p. It has solutions p = ` and p = (` + 1).
Hence the general solution for R is
B
R = Ar` + `+1 .
r
and so the solution for U is


B
`
U(r, ) = Ar + `+1 P` (cos ).
r
Because any linear combination of solutions of the equation is also a solution of the equation ( 2 is a linear
operator), the general solution is a sum of these, i.e:


Bn
U(r, ) = An rn + n+1 Pn (cos ).
(7.12)
r
n=0
101

The individual functions on the right are axisymmetric spherical harmonics and they form a set of axisymmetric solutions of Laplaces equation which is complete, i.e. (7.12) can be shown to be the most general
axisymmetric solution.
One can match arbitrary boundary conditions to an infinite series of Legendre polynomials using their
orthogonality properties (see later). However, in this course we will stick to problems where only a few terms
are needed and we can see what they are by intelligent guesswork: the essential rule is only to put into the
prospective answer those Legendre polynomials which appear in the boundary conditions.
Example 7.5. A perfectly spherical conductor, centre 0, radius a, is placed in an otherwise uniform electric field E0 . (Mathematically, the condition for a conductor is that the electrostatic potential U is constant.)
What is the potential everywhere outside the conductor? And inside?
Outside the conductor (r > a), we want to solve 2U = 0. The boundary conditions are that U =constant
on r = a and that far from the conductor U E0 .
The unperturbed field (the one before the conductor was added) is E = E 0 k, choosing the z-axis to align
with the field. Converting this to the es of spherical polars, we have
E0 = E0 cos er E0 sin e

which is what the field must look like as r : this has potential

U0 = E0 r cos + constant = E0 rP1 (cos ) + constant.

(Note that this is a solution of Laplaces equation.) Now our potential




Bn
U = An rn + n+1 Pn (cos ) An rn Pn (cos )
r
n=0
n=0
as r . But this must equal U0 = E0 rP1 (cos )+const. at large r, so we can deduce that A1 = E0 , A0 is an
arbitrary constant, and An = 0 for all other n.
On r = a we want U to be constant, i.e. it should not vary with . Now on r = a



Bn
B1
B0
+ E0 a + 2 P1 (cos ) + n+1 Pn (cos )
U(a, ) = A0 +
a
a
n=2 a
The potential on r = a will vary with unless all the coefficients of Pn (cos ) (n > 0) each vanish. Hence we
must have B1 = E0 a3 to make the bracket vanish, and Bn = 0 (n 2). Hence finally the solution is


B0
a3
U(r, ) = A0 +
+ E0 r 2 cos .
r
r
Note that A0 and B0 are undetermined constants. To determine B0 we need additional information to ascertain
the potential difference between the surface of the conductor and a point at infinity. The constant A 0 will
always be arbitrary, because the absolute value of the potential has no physical meaning (only its gradient is
actually observable).
Inside, since U is constant on the boundary, it must be constant inside the conductor.
This last point has practical consequences. The voltage in space [in a static field] satisfies Laplaces
equation. If you stand under an electricity pylon, there is a rather large voltage changethousands of volts
between your head and your feet. But if you stand inside a wire cage (often called a Faraday cage), then
the wire acts like a continuous conductor and equalizes the voltage over the cage and hence inside the cage
too. That is why a wire cage provides a refuge from lightning. Cages also provide screening from electronic
surveillance, or, by putting equipment inside them, safety for the people outside.
102

Exercise 7.3. Show that at a general point the following are solutions of Laplaces equation 2U = 0.
1. U = rn cosn , for an integer n, in cylindrical polar coordinates.
2. U = r sin cos , in spherical polar coordinates.
2

7.6 Introduction to Legendre polynomials

We now take a brief look at the Legendre polynomials. These are defined as the solutions of Legendres
differential equation which is


d
df
(1 x2)
+ f = 0.
dx
dx
or similar, where is an arbitrary constant. The solution of this is outside the scope of this course, but briefly
we search for power-law solutions of the form
f (x) = a p x p
. Then, it can be shown that the series only converges at both x = 1 if = `(` + 1) where ` is an integer,
and we can take ` as a non-negative integer without loss of generality.
Then, the function f (x) which satisfies the above D.E. for = `(`+1) is called the Legendre polynomial
of degree `, usually written P` (x). (It is common to use letter ` for this subscript, since when things are
extended to 3-D spherical harmonics, letters n and m are generally used for other functions in the r and
coordinates.)
There is an arbitrary multiplicative constant in each P` , which is chosen so that P` (1) = 1 for all `. It turns
out that P` is an `-th order polynomial, and involves only even/odd powers of w if ` is even/odd.
The solutions can be obtained by Rodrigues formula
P` (x) =

2` `!

d` 2
[(x 1)` ]
dx`

There is also a recurrence relation between them,

P`+1(x) =

1
[(2` + 1) xP`(x) `P`1(x)]
`+1

which gives all of them, working upwards from P0 and P1 .

Starting from Rodriguess formula
P0 (x) = 1
P1 (x) = x
then the recurrence relation gives subsequent ones as
P2 (x)

1 2
(3x 1)
2
103

(7.13)

P3 (x)
P4 (x)

1 3
(5x 3x)
2
1
=
(35x4 30x2 + 3)
8
etc
=

Another important property is orthogonality, i.e. the fact that

Z 1

Pm (w)Pn (w) dw = 0 if m 6= n
=

2
2n + 1

if m = n

This property enables us to express any general function as an infinite series of Legendre polynomials, by a
device similar to that for calculating Fourier coefficients.
In this course we will only look at simple functions, in which case a general n-th order polynomial can be
rearranged into a sum of the first n Legendre polynomials, e.g. suppose we are given a boundary condition
in Laplaces equation looking like f (w) = w2 + w + 1, (w cos ) we need to choose a sum of Legendre
polynomials to match this. We need (2/3)P2(w) to match the quadratic w2 term. Then we need 1P1 (w) to
match the linear term. Finally for the constant, the (2/3)P2 has given us a 1/3 constant term, so to match
the 1 we need +(4/3)P0 on the right hand side. So in that example,
2
4
w2 + w + 1 P2 (w) + 1P1(w) + P0 (w)
3
3
and we can now put the right-hand-side into the general solution to Laplaces equation, 7.12 , and choose
suitable constants to match.

104