Chapter 6

The Laplace Equation

Many of the laws of Physics can be formulated as partial dierential equations that involve the
Laplacian or del squared operator, which in Cartesian coordinates is dened as

2
=

2
x
2
+

2
y
2
+

2
z
2
. (6.1)
We have already seen the one-dimensional version of an important example, the heat equation,
in Sec. 2.3. In three dimensions this is
u
t
=
2
u + q(x, y, z, t) . (6.2)
If we consider a system in steady-state equilibrium, then the temperature u is no longer changing
as a function of time and the term u/t is zero. Furthermore if the system has no internal heat
sources then we have q = 0 and the heat equation becomes

2
u = 0 . (6.3)
This is called the Laplace equation, and we will focus on solving it in this chapter. Before doing
that, however, we will mention several other important equations that involve the Laplacian
operator. Another example is the wave equation,

2
u
1
v
2

2
u
t
2
= 0 , (6.4)
where v here corresponds to with which the waves propagate through space. Another closely
related example is the Schrodinger equation for the wavefunction of a particle of mass m in
a potential V ,

h
2
2m

2
+ V = ih

t
. (6.5)
In some problems this reduces to the Helmholtz equation,
63
64 Lecture Notes on Mathematical Methods

2
+ k
2
= 0 , (6.6)
where for example the solution may be proportional to the spatial part of a wavefunction.
Maxwells equations for the electric and magnetic elds are another important example.
These are a system of coupled partial dierential equations for E and B that are rst order in
both space and time. The corresponding equations for the scalar and vector electromagnetic
potentials are second order. For example, the Poisson equation describes the electrostatic
potential u in a system with charge density ,

2
u =

0
. (6.7)
An important special case is when the charge density is zero, and then Eq. (6.7) reduces to the
Laplace equation (6.3).
The Laplace equation is the simplest of these examples and we solve it in this chapter
using separation of variables. Much of what we will encounter, however, is applicable to the
other equations above as well. We will nd that, depending on the symmetry of the boundary
conditions, the problem may simplify when it is expressed in a particular coordinate system. For
example, if we try to nd the electrostatic potential in a rectangular region, then we will nd
that separation of variables leads to equations that can be easily solved when using Cartesian
coordinates. We will see an example of this in Sec. 6.1.
If the problem possesses, however, a spherical symmetry, then it is easier to solve in spherical
coordinates. In Sec. 6.2 we will see, therefore, how to express the equations mentioned above
in several dierent coordinate systems. When we apply the method of separation of variables
to these equations when they are expressed in, say, spherical (r, , ) or cylindrical (, , z)
coordinates, we are led to ordinary dierential equations involving these variables. The solutions
to these equations are referred to as the special functions of mathematical physics, such as
Legendre polynomials, Bessel functions, Hermite polynomials, etc., and we will continue to study
these throughout the course.
6.1 The Laplace equation in Cartesian coordinates
The Laplace equation in Cartesian coordinates is a partial dierential equation for the three
spatial variables x, y and z:

2
u
x
2
+

2
u
x
2
+

2
u
x
2
= 0 . (6.8)
In this section we will solve the Laplace equation subject to boundary conditions that are
easy to express in terms of x, y and z. Consider a region of square cross section that extends
innitely far in the z directions. Suppose the sides at x = 0, x = L and y = 0 are held at
a potential of u = 0, and the potential on the face at y = L has a nonzero value given by a
function of x, i.e., u(x, L, z) = f(x).
Since the system has no variation in z we must have u/z = 0, and so the Laplace equation
for this problem really only involves the variables x and y, i.e., we need to solve
The Laplace Equation 65

2
u
x
2
+

2
u
y
2
= 0 . (6.9)
The boundary conditions for the solution u(x, y) are
u(0, y) = 0 , (6.10)
u(L, y) = 0 , (6.11)
u(x, 0) = 0 , (6.12)
u(x, L) = f(x) , (6.13)
where the function f(x) needs to be supplied; it is analogous to the initial condition that was
specied for the heated-rod problem of Sec. 2.3. For now we will leave open the option to choose
any f(x) as long as is consistent with the other boundary conditions imposed on u, i.e., we
require f(0) = f(L) = 0. The basic geometry of the system and boundary conditions are shown
in Fig. 6.1.
Figure 6.1: Cross section of an
innitely long rectangular region with
sides held at the potentials indicated
(see text).
The situation is somewhat dierent from that of the heated rod in that here we have one set
of boundary conditions for x that are homogeneous, i.e., the right-hand sides of Eqs. (6.10) and
(6.11) are zero. But the boundary conditions for y involve one equation, u(x, L) = f(x), where
the right-hand side is not zero. We will see why this is relevant in a short while.
6.1.1 Separation of variables in Cartesian coordinates
We begin the same way as before by seeking product solutions (x, y) to the two-dimensional
Laplace equation (6.9) of the form
(x, y) = X(x)Y (y) . (6.14)
As before we will nd a family of such solutions
n
and we will construct the nal u(x, y) from a
linear combination of them. We will insist that they satisfy those boundary conditions that are
homogeneous, i.e., (0, y) = (L, y) = (x, 0) = 0. But we cannot impose the nonhomogeneous
boundary condition (x, L) = f(x), since a linear combination of the form

n
c
n

n
(x, L) would
not be equal to f(x), and thus the nal answer u(x, y) would not satisfy Eq. (6.13).
66 Lecture Notes on Mathematical Methods
The two-dimensional Laplace equation for = XY is

2
= XY

+ X

Y = 0 . (6.15)
Dividing both sides by XY gives
X

X
+
Y

Y
= 0 . (6.16)
As the two terms on the left are functions of x alone and y alone, they must each be constant
and they must sum to zero. That is, we must have
X

X
= , (6.17)
Y

Y
= , (6.18)
where is a separation constant. Notice that (6.17) can be written
LX = X (6.19)
with L = d
2
/dx
2
. Comparison with Eq. (5.2) shows that this is a special case of the Sturm-
Liouville equation with
p(x) = 1 , (6.20)
q(x) = 0 , (6.21)
w(x) = 1 , (6.22)
and with the eigenvalue of the SL equation taken as the negative of our separation constant . (In
the following we will, following common usage, still refer to as the eigenvalue.) Furthermore,
the boundary conditions for X, Eqs. (6.10) and (6.11) are homogeneous, so therefore L is a self-
adjoint operator. The solutions X will thus have the properties we discussed in Chapter 5
(orthogonality, completeness, etc.).
We cannot say the same for the equation for Y , however, even though the equation is the
same apart from a minus sign. This is because the boundary conditions for (6.18) are not
homogeneous, and therefore the operator (which implicitly includes the boundary conditions) is
not self-adjoint.
At this point the problem for X looks the same as the heated-rod of Sec. 2.3. As done there,
we can see if solutions exist for = 0 or > 0, and in both cases we will nd that these only
lead to the trivial solution, i.e., u = 0. For < 0 we dene = k
2
for real k, and nd the
solution to the X equation
X = Acos(kx) + B sin(kx) . (6.23)
The Laplace Equation 67
Applying the boundary condition (0, y) = 0 gives
(0, y) = X(0)Y (y) = Y (y)(A + 0) = 0 , (6.24)
and therefore A = 0. Requiring (L, y) = 0 gives
(L, y) = X(L)Y (y) = Y (y)B sin(kx) = 0 , (6.25)
which we can satisfy if k = n/L for integer n. As before we do not consider n = 0, as this gives
the trivial solution, nor n < 0, as these are simply the negatives of the corresponding solutions
with n > 0. We will again use a subscript n to label the solution and its eigenvalue, e.g.,

n
= k
2
n
=
_
n
L
_
2
. (6.26)
We now turn to the equation for Y ,
Y

k
2
Y = 0 , (6.27)
Using the methods of Chapter 1 we can write down the general solution as
Y (y) = Ae
ky
+ Be
ky
. (6.28)
Applying the nal homogeneous boundary condition (x, 0) = 0 gives
(x, 0) = X(x)Y (0) = X(x)(A + B) = 0 , (6.29)
and so we must have A = B. We can therefore write the solution A(e
ky
e
ky
) = 2Asinh(ky)
in an equivalent form (with redenition of the new A as two times the old A):
Y = Asinh(ky) . (6.30)
The product solution for the eigenvalue
n
= k
2
n
= (n/L)
2
is therefore

n
(x, y) = sinh(k
n
y) sin(k
n
x) . (6.31)
Any linear combination of the
n
will satisfy the three homogeneous boundary conditions,
but not the nonhomogeneous one u(x, L) = f(x). But we can form a linear combination of the

n
,
u(x, y) =

n=1
b
n
sinh
_
ny
L
_
sin
_
nx
L
_
, (6.32)
and try to nd values of the coecients b
n
so that the nal boundary condition is satised. That
is, we need to determine the b
n
such that
68 Lecture Notes on Mathematical Methods
f(x) = u(x, L) =

n=1
b
n
sinh(n) sin
_
nx
L
_
. (6.33)
To nd the coecients we employ exactly the same trick as with the heated-rod problem. That
is, we notice that the product solutions contain factors of sin(nx/L), and we have seen in
the previous chapter that these form a complete set of functions in the interval [0, L], so any
f(x) with f(0) = f(L) = 0 can be expressed as a linear combination of sines. (In problems with
dierent boundary conditions we could have sines or cosines; the cosines are absent here because
we imposed u(0, y) = 0.)
If we therefore multiply both sides of Eq. (6.33) by sin(mx/L) and integrate from 0 to L
we nd
_
L
0
f(x) sin
_
mx
L
_
dx =

n=1
b
n
sinh(n)
_
L
0
sin
_
nx
L
_
sin
_
mx
L
_
dx
=

n=1
b
n
sinh(n)
L
2

nm
=
b
m
L
2
sinh(m) . (6.34)
The coecients b
n
are therefore
b
n
=
2
Lsinh(n)
_
L
0
f(x) sin
_
nx
L
_
dx , (6.35)
and these together with Eq. (6.32) give the nal answer u(x, y).
6.1.2 Examples
As an example, suppose f(x) = u
0
where u
0
is a constant with units of electrical potential. That
is, three of the surfaces are at ground potential and the fourth is held at u
0
. The coecients b
n
are
b
n
=
2
Lsinh(n)
_
L
0
u
0
sin
_
nx
L
_
dx =
2u
0
n sinh(n)
[1 cos(n)] , (6.36)
which can also be written as
b
n
=
_

_
0 n even,
4u
0
n sinh(n)
n odd.
(6.37)
This solution is shown in Fig. 6.2. The same problem can be solved numerically using the
relaxation method as described in Appendix A.
The Laplace Equation 69
x
0
1
y
0
1
u
(
x
,
y
)
0
1
Figure 6.2: A plot of the potential
u(x, y) based on the solution from
Eqs. (6.32) and (6.37) (see text).
6.1.3 Case of all sides with nonhomogeneous boundary conditions
The problem we have worked out above required that three of the four sides have homogeneous
boundary conditions, which is to say the potential for those sides was set to zero. This may
seem so restrictive that one might think the approach is not useful in practice. In fact we can
easily modify the calculation to cover the case where all four of the sides have nonhomogeneous
boundary conditions. Let us suppose the boundary conditions are given by
u(0, y) = f
a
(y) , (6.38)
u(x, L) = f
b
(x) , (6.39)
u(L, y) = f
c
(y) , (6.40)
u(x, 0) = f
d
(x) , (6.41)
for some specied functions f
a
(y), f
b
(x), f
c
(y) and f
d
(x). Our goal is to solve
2
u(x, y) = 0
subject to Eqs. (6.38) (6.41).
To do this, rst consider the related problem where we impose only one of the four boundary
conditions above and we make the other three homogeneous. That is, let us solve

2
u
a
= 0 (6.42)
subject to the boundary conditions
u(0, y) = f
a
(y) , (6.43)
u(x, L) = 0 , (6.44)
u(L, y) = 0 , (6.45)
u(x, 0) = 0 . (6.46)
70 Lecture Notes on Mathematical Methods
We can nd the solution u
a
with the method described above in Sec. 6.1.1. Then we can carry
out the corresponding procedure for the other faces, e.g., we make all of the boundary conditions
except (6.39) homogeneous, so that the only nonhomogeneous term is f
b
(x) and thus obtain the
solution u
b
, and in a similar way we also nd u
c
and u
d
.
We now claim that the solution to the full problem with the boundary conditions given by
Eqs. (6.38) (6.41) is simply the sum
u = u
a
+ u
b
+ u
c
+ u
d
. (6.47)
It is easy to verify that this works. Applying the Laplacian operator to Eq. (6.47) gives

2
u =
2
(u
a
+ u
b
+ u
c
+ u
d
) =
2
u
a
+
2
u
b
+
2
u
c
+
2
u
d
= 0 , (6.48)
because all of the terms u
a
, u
b
, u
c
and u
d
are individually solutions to the Laplace equation.
Furthermore we can check if the boundary conditions hold. For example, for the left-hand
face at x = 0 we have
u(0, y) = u
a
(0, y) + u
b
(0, y) + u
c
(0, y) + u
c
(0, y)
= f
a
(y) + 0 + 0 + 0 . (6.49)
That is, the rst term on the right-hand side is u
a
(0, y) = f
a
(y). For the other three terms
corresponding to u
b
, u
c
and u
d
, the boundary condition for face at x = 0 is that the potential is
zero. So the boundary conditions for the full u(x, y) hold at x = 0 and one can easily show in
the same manner that they are satised on the other three sides as well.
6.1.4 Why it worked in this case and wont in others
It is important now to look back at what we have just done and think about the logic of the
various steps. Notice that the assumption of a product solution of the form = X(x)Y (y) will
always lead to the separated equations (6.17) and (6.18) regardless of the boundary conditions.
But then we were lucky in that the two boundary conditions u(0, y) = 0 and u(L, y) = 0 led to an
eigenvalue equation for X and the eigenfunctions sin(nx/L) formed a complete set of functions
in the interval [0, L]. Therefore we could use these to match the remaining nonhomogeneous
boundary condition u(x, L) = f(x) for arbitrary f(x).
It is easy to construct a problem that would not work out so nicely. Suppose, for example,
that the region in question had been triangular, as shown in Fig. 6.3, with the boundary
conditions u(0, y) = u(x, 0) = 0 and u(x, L x) = f(x).
If we follow through the same steps as before we nd, after separating variables and applying
the rst two homogeneous boundary conditions, product solutions of the form
(x, y) = sin(kx) sinh(ky) . (6.50)
The Laplace Equation 71
Figure 6.3: Cross section in x, y of
a triangular region innitely long in
z with sides held at the potentials
indicated (see text).
But now there is no boundary condition like u(0, L) = 0, which is what restricted k to take
on discrete values. We are still left with trying to satisfy u(x, L x) = f(x). This is a
nonhomogeneous boundary condition and as we do not have the type of constraint that limits
the values of k to the discrete values (n/L), we do not obtain from the sin(kx) terms an
orthogonal, complete set of eigenfunctions. Instead of a sum over discrete values of k we could
try a solution of the form
u(x, y) =
_
b(k) sin(kx) sinh(ky) dk , (6.51)
where b(k) is a function that we need to determine so that we satisfy the nal boundary condition.
One can easily verify that Eq. (6.51) is indeed a solution to our PDE and it satises the rst
two boundary conditions. We would then need to nd the function b(k) such that it satises
the nal condition, i.e.,
f(x) =
_
b(k) sin(kx) sinh(k(L x)) dk . (6.52)
But there is no corresponding orthogonality relation for the functions sin(kx) sinh(k(L x)),
not to mention the fact that they appear here in an integral, rather than the sum (6.33) that we
found above involving sin(nx/L). So we have no simple way of nding the b(k) that satises
all of the boundary conditions.
For real-world problems, the boundary conditions are often so complicated that the only
possible way to solve them is to use numerical methods. There is more information on this in
Appendix A. If the boundary conditions exhibit a particular symmetry, however, one can still
nd a solution in closed form if the problem is expressed in a particular coordinate system. We
explore this option in the following section.
6.2 The Laplace equation in 2D polar coordinates
Suppose we want to nd the electrical potential inside a circular tube of radius a which extends
innitely far in the z direction, as shown in Fig. 6.4.
Suppose the outer edge of the tube, i.e., at a radius r = a, is held at a potential given by a
function f(), where is the angle in polar coordinates as indicated in Fig. 6.4.
72 Lecture Notes on Mathematical Methods
Figure 6.4: Cross section of an
innitely long circular region with
radius a and the outer edge held at a
potential u(a, ) = f() (see text).
To nd the potential u everywhere inside the tube we need to solve the Laplace equation
subject to the specied boundary conditions. As the system does not have any dependence on
the coordinate z we have u/z = 0 and therefore we are again dealing with the two-dimensional
Laplace equation,

2
u
x
2
+

2
u
y
2
= 0 . (6.53)
Because of the circular symmetry of the boundary conditions, however, we will nd it easiest to
solve the problem if it is expressed in polar coordinates, i.e.,
r =
_
x
2
+ y
2
, (6.54)
= tan
1
(y/x) , (6.55)
and we will write the solution as u(r, ). The boundary condition may thus be written as
u(a, ) = f() . (6.56)
The angle is of course the same as +2n for any integer n, so the function f() must satisfy
f() = f( + 2n). The solution u(r, ) must also satisfy this, i.e.,
u(r, ) = u(r, + 2n) . (6.57)
A nal requirement is that the function should be nite everywhere in the interior of the circle,
i.e.,
|u(r, )| < . (6.58)
6.2.1 The Laplacian in 2D polar coordinates
We now need to express the Laplace equation (6.53), which involves the derivatives
2
u/x
2
and
2
u/y
2
, in terms of r and . This is a somewhat tedious but straightforward exercise in
The Laplace Equation 73
the application of the chain and product rules of dierentiation. We start by using the chain
rule to nd
u
x
=
u
r
r
x
+
u

x
. (6.59)
Using the product rule on each of the two terms to dierentiate one more time gives

2
u
x
2
=
u
r

2
r
x
2
+
r
x

x
_
u
r
_
+
u

x
2
+

x

x
_
u

_
. (6.60)
We now need to use the chain rule again to nd

x
_
u
r
_
=
r
x

2
u
r
2
+

x

2
u
r
, (6.61)

x
_
u

_
=
r
x

2
u
r
+

x

2
u

2
. (6.62)
Substituting these into Eq. (6.60) gives

2
u
x
2
=

2
r
x
2
u
r
+
r
x
_
r
x

2
u
r
2
+

x

2
u
r
_
+

2

x
2
u

+

x
_
r
x

2
u
r
+

x

2
u

2
_
. (6.63)
Now using Eqs. (6.54) and (6.55), which dene the relation between the two coordinate systems,
we can nd the required derivatives:
r
x
=
x
_
x
2
+ y
2
=
x
r
, (6.64)

x
=
y
x
2
+ y
2
=
y
r
2
, (6.65)

2
r
x
2
=
y
2
(x
2
+ y
2
)
3/2
=
y
2
r
3
, (6.66)

x
2
=
2xy
(x
2
+ y
2
)
2
=
2xy
r
4
. (6.67)
Our goal is to express the result for the Laplacian in terms of r and , not containing x and y as
we have above. Rather than making the replacements x = r cos and y = r sin now, however,
we will see below that the x and y terms that appear on the right-hand sides above will either
cancel or appear in the form x
2
+y
2
, which we can replace with r
2
. Substituting the derivatives
above into Eq. (6.63) we nd
74 Lecture Notes on Mathematical Methods

2
u
x
2
=
y
2
r
3
u
r
+
x
2
r
2

2
u
r
2

2xy
r
3

2
u
r
+
2xy
r
4
u

+
y
2
r
4

2
u

2
. (6.68)
By going through the same steps again for
2
u/y
2
we nd

2
u
y
2
=
x
2
r
3
u
r
+
y
2
r
2

2
u
r
2
+
2xy
r
3

2
u
r

2xy
r
4
u

+
x
2
r
4

2
u

2
. (6.69)
Adding these terms together we nd the two-dimensional Laplacian operator in polar coordinates
is

2
u
x
2
+

2
u
y
2
=

2
u
r
2
+
1
r
u
r
+
1
r
2

2
u

2
, (6.70)
which is often written in the equivalent form

2
u
x
2
+

2
u
y
2
=
1
r

r
_
r
u
r
_
+
1
r
2

2
u

2
. (6.71)
6.2.2 Separation of variables in polar coordinates
We can nally return to our problem of solving the two-dimensional Laplace equation in the
circular region shown in Fig. 6.4. This time we seek product solutions of the form
(r, ) = R(r)Q() , (6.72)
where R(r) only depends on r and Q() only depends on . Using the Laplace equation in polar
coordinates (6.70) we nd
R

Q +
1
r
R

Q +
1
r
2
RQ

= 0 . (6.73)
Multiplying both sides by r
2
/RQ gives
r
2
R

R
+
rR

R
+
Q

Q
= 0 . (6.74)
The rst two terms depend only on r and the third only on , so we must have
r
2
R

R
+
rR

R
=
Q

Q
= , (6.75)
where is a separation constant. Our problem therefore reduces to the two ordinary dierential
equations
r
2
R

+ rR

R = 0 , (6.76)
Q

+Q = 0 . (6.77)
The Laplace Equation 75
The next step is to investigate for what values of we can nd solutions. If < 0, i.e.,
=
2
for real, we have
Q

2
Q = 0 , (6.78)
which has the solution
Q() = Ae

+ Be

, (6.79)
where A and B are arbitrary constants. If we now impose the periodicity requirement (6.57) we
have Q() = Q( + 2n) for any integer n. Using this with = 0 in Eq. (6.79) gives
A + B = Ae
2n
+ B
2n
. (6.80)
This must hold for arbitrary n, so if we consider n we get
A + B = A+ B , (6.81)
which can only hold if A = 0. Equation (6.80) therefore becomes
B = Be
2n
. (6.82)
This can only be satised if B = 0 or = 0. But we started by assuming < 0, so therefore
this leads to A = B = 0, that is, the trivial solution u = 0.
If we now consider = 0, then the equation for Q becomes simply Q

= 0, which has the

solution
Q = A + B . (6.83)
If we now require that Q() = Q( + 2) we nd
A + B = A + B( + 2) . (6.84)
This is only possible if B = 0 but then works for arbitrary A. So Q = A is a solution that
satises the periodic boundary condition for arbitrary A. We still need to impose the boundary
condition u(a, ) = f(), which we will investigate shortly.
Finally if we consider > 0, i.e., =
2
for real, the dierential equation for Q becomes
Q

+
2
Q = 0 , (6.85)
which has the solution
Q() = Acos() + B sin() . (6.86)
76 Lecture Notes on Mathematical Methods
This must give the same value if we advance by 2. Therefore if we increase by 2, the
arguments of the sin and cos functions must change by an integral multiple of 2, i.e., we must
have
( + 2) = + 2n (6.87)
for any integer n. Solving Eq (6.87) for we nd simply = n, i.e., = n
2
.
We can therefore write the Eq. (6.76) for R as
r
2
R

+ rR

n
2
R = 0 . (6.88)
This is an example of the Euler equation, which we will encounter again in Sec. 7.2.2. It
is a second-order dierential equation with non-constant coecients, so we cannot solve it
directly using the methods of Chapter 1. In this case, however, we can employ a clever variable
substitution by dening
z = ln(r/a) , (6.89)
where a is the radius of the circular region. We can convert Eq. (6.88) into an equation for z by
rst working out the required derivatives:
R

=
dR
dr
=
dR
dz
dz
dr
=
dR
dz
1
r
, (6.90)
R

=
d
dr
_
dR
dz
1
r
_
=
1
r
2
dR
dz
+
1
r
2
d
2
R
dz
2
. (6.91)
Substituting these into Eq. (6.88) gives the dierential equation for R(z),
d
2
R
dz
2
n
2
R = 0 , (6.92)
which now has constant coecients and can be easily solved. Its solution is
R(z) = Ae
nz
+ Be
nz
, n = 1, 2, . . . . (6.93)
Converting this back into a function of r gives
R(r) = A
_
r
a
_
n
+ B
_
r
a
_
n
n = 1, 2, . . . . (6.94)
Our nal boundary condition, Eq. (6.58), states that the solution should be nite everywhere
inside the circular region. But when r 0, i.e., in the middle of the circle, the solution diverges
unless B = 0 for both the n = 0 and n = 1, 2, . . . cases. The solution for R therefore must have
the form
The Laplace Equation 77
R(r) = A
_
r
a
_
n
, n = 0, 1, 2, . . . . (6.95)
Note that this solution covers both cases n = 0 (i.e., = 0, where we found a constant solution)
and n = 1, 2, . . ..
Combining this with the solution to the equation for Q gives the product solutions, which
label with a subscript n to indicate the value of the separation constant ( = n
2
) is

n
(r, ) =
_
r
a
_
n
[A
n
cos(n) + B
n
sin(n)] . (6.96)
We still need to satisfy the boundary condition u(a, ) = f(), and to do this we will construct
the nal answer out of a linear combination of the product solutions,
u(r, ) = A
0
+

n=1
_
r
a
_
n
[A
n
cos(n) + B
n
sin(n)] . (6.97)
We then determine the coecients A
n
and B
n
so that the boundary condition is satised, i.e.,
we require
f() = u(a, ) = A
0
+

n=1
[A
n
cos(n) + B
n
sin(n)] . (6.98)
To nd the coecients we can again exploit the orthogonality of the sin(n) and cos(n) terms.
For example, if we multiply both sides of Eq. (6.98) by cos(m) for arbitrary integer m and then
integrate from 0 to 2 we nd
_
2
0
f() cos(m) d =
_
2
0
A
0
cos(m) d
+

n=1
A
n
_
2
0
cos(m) cos(n) d
+

n=1
B
n
_
2
0
cos(m) sin(n) d . (6.99)
The rst integral in Eq. (6.99) is easily shown to be zero. For the other integrals we can exploit
the orthogonality relations
_
2
0
sin(m) sin(n) d =
mn
, (6.100)
_
2
0
cos(m) cos(n) d =
mn
, (6.101)
_
2
0
sin(m) cos(n) d = 0 , (6.102)
78 Lecture Notes on Mathematical Methods
These are similar to the relations we saw in Eqs. (2.47) in connection with the heated-rod
problem but with an interesting dierence. There the region of integration for the variable x
was [0, L], so the region of integration corresponded to only a half-period of the sine function or
integral multiples thereof, i.e., the argument of the function sin(x/L) went from 0 to 0 to .
This was possible because in that problem the eigenfunctions only included sin(nx/L), whereas
here we have both sines and cosines. But while the rst two relations above, (6.100) and (6.101),
would hold even if we integrate from 0 to , the third relation relation (6.102) that involves both
sine and cosine factors only holds for arbitrary n and m if we integrate over a full period, i.e.,
0 to 2.
The integral in Eq. (6.99) with sin(m) cos(n) is therefore zero for all n, and the integral
with cos(m) cos(n) is only nonzero for n = m. The coecient A
n
is therefore
A
n
=
1

_
2
0
f() cos(n) d . (6.103)
in a similar way we nd the coecients B
n
to be
B
n
=
1

_
2
0
f() sin(n) d . (6.104)
and the term A
0
is
A
0
=
1
2
_
2
0
f() d . (6.105)
Equations (6.103) (6.105) for the coecients together with Eq. (6.97) therefore represent the
nal answer for u(r, ).
6.2.3 Examples
Suppose, for example, the potential on the edge of circular region is given by a constant, i.e.,
f() = u
0
. We can then use Eqs. (6.103) and (6.104) to nd the coecients A
n
and B
n
. The
integrals are easy to carry out and one nds A
n
= B
n
= 0 for all n = 1, 2, . . .. We then use
Eq. (6.105) to nd the coecients A
0
,
A
0
=
1
2
_
2
0
u
0
d = u
0
. (6.106)
Thus we nd
u(r, ) = u
0
, (6.107)
that is, the potential everywhere in the circle is the same as the value on the edge. This is an
example of a general property of the Laplace equation, namely, that the extreme values of u
always appear on a boundary, never in the interior. And since here the boundary was held at
u(a, ) = u
0
, one must nd u(r, ) = u
0
everywhere in the circle.
The Laplace Equation 79
Now suppose that the potential on the edge is given by f() = u
0
sin(2). Using the
orthogonality relations (6.100) (6.102) one easily nds that all of the coecients are zero
except B
2
, which is
B
2
=
1

_
2
0
u
0
sin
2
(2) d = u
0
. (6.108)
The potential as a function of r and is therefore
u(r, ) = u
0
_
r
a
_
2
sin(2) . (6.109)
Notice that because of the (r/a)
2
term, the dependence of u on angle becomes washed out as
one moves towards the centre of the circle, as can be seen in Fig. 6.5.
Figure 6.5: The solution to the
Laplace equation shown in a cross
section of a circular tube as found
from Eq. (6.109). The potential is
proportional to the plotted density.
x
y
6.3 The Laplacian in cylindrical and spherical polar coordinates
Finally in this chapter we extend our discussion of dierent coordinate systems to cylindrical
and spherical polar coordinates. In the previous sections we have seen that when we carry out
separation of variables in dierent coordinate systems, we are led to dierent sets of ordinary
dierential equations. For example, separating the Laplace equation in a two-dimensional polar
coordinates led to the Euler equation (6.88) for the radial coordinate.
When we use separation of variables for many of the other important partial dierential
equations often encountered in physics (Schrodinger, Helmholtz, wave, heat equations, etc.) in
cylindrical or spherical polar coordinates, the resulting ordinary dierential equations, as with
the Euler equation, often do not have constant coecients. Unlike the Euler equation, however,
there is usually no simple change of variables that will convert them into one with constant
coecients. We must then resort to other methods such as series solution and the Frobenius
method, which we will return to in Chapter 7.
For now, however, we will simply dene cylindrical and spherical polar coordinates and
provide the form of the Laplacian in these two systems, and we will make use of these later on
in the course.
80 Lecture Notes on Mathematical Methods
6.3.1 Cylindrical coordinates
If the boundary conditions possess a cylindrical symmetry it will be easiest to solve the problem if
it is expressed in cylindrical coordinates (, , z), which are related to the Cartesian coordinates
(x, y, z) through
x = cos , (6.110)
y = sin , (6.111)
z = z . (6.112)
The coordinates are dened such that 0 < , 0 2 and < z < . The inverse
relations are
=
_
x
2
+ y
2
, (6.113)
= tan
1
_
y
x
_
, (6.114)
z = z . (6.115)
The relation between cylindrical and Cartesian coordinates is illustrated in Fig. 6.6. Of course
the quantities and are essentially the same as the r and we used in Sec. 6.2 as two-
dimensional polar coordinates, but with three dimensions we often use the standard notation
, , z.
Figure 6.6: Illustration of cylindrical
coordinates (see text).
The Laplacian in cylindrical coordinates is easily found from the expression we derived in
Sec. 6.2.1) for polar coordinates, since the missing term
2
u/z
2
is the same as in Cartesian
coordinates. The Laplacian operating on a function u(, , z) is therefore

2
u =

2
u

2
+
1

+
1

2
u

2
+

2
u
z
2
. (6.116)
For reference we also give here the volume element dV in cylindrical coordinates:
The Laplace Equation 81
dV = d ddz . (6.117)
(6.118)
This is illustrated schematically in Fig. 6.7.
Figure 6.7: Illustration of the volume
element dV in cylindrical coordinates
(see text).
6.3.2 Spherical polar coordinates
If the boundary conditions of the problem have a spherical symmetry, e.g., they are specied in
terms of the distance of a boundary from the origin, then the problem may be easier to solve in
spherical polar coordinates (r, , ). These are related to Cartesian coordinates by
x = r sin cos , (6.119)
y = r sin sin , (6.120)
z = r cos , (6.121)
where 0 2, 0 and r 0. The inverse relations are
r =
_
x
2
+ y
2
+ z
2
, (6.122)
= tan
1
z
_
x
2
+ y
2
, (6.123)
= tan
1
_
y
x
_
. (6.124)
The relation between Cartesian and spherical polar coordinates is illustrated in Fig. 6.8.
The Laplacian operator can be translated in to spherical coordinates using methods similar
to those seen in Sec. 6.2.1 for the case of plane polar coordinates. The calculation is rather long
and so we will simply quote the result. The Laplacian of a function u(r, , ) in spherical polar
coordinates is
82 Lecture Notes on Mathematical Methods
Figure 6.8: Illustration of spherical
coordinates (see text).

2
u =
1
r
2
sin
_
sin

r
_
r
2
u
r
_
+

_
sin
u

_
+
1
sin

2
u

2
_
. (6.125)
For reference we also give here the area, volume and solid angle elements in spherical polar
coordinates:
dA = r
2
sin d d , (6.126)
dV = dAdr = r
2
sin d ddr , (6.127)
d =
dA
r
2
= sin d d . (6.128)
These relations are illustrated in Fig. 6.9.
(a) (b)
Figure 6.9: Illustration of (a) the area and solid angle and (b) volume elements in spherical coordinates
(see text).