Part IB Methods
Based on lectures by D. B. Skinner
Notes taken by Dexter Chua
Michaelmas 2015
These notes are not endorsed by the lecturers, and I have modified them (often
significantly) after lectures. They are nowhere near accurate representations of what
was actually lectured, and in particular, all errors are almost surely mine.
Self-adjoint ODEs
Periodic functions. Fourier series: definition and simple properties; Parseval’s theorem.
Equations of second order. Self-adjoint differential operators. The Sturm-Liouville
equation; eigenfunctions and eigenvalues; reality of eigenvalues and orthogonality of
eigenfunctions; eigenfunction expansions (Fourier series as prototype), approximation
in mean square, statement of completeness. [5]
PDEs on bounded domains: separation of variables
Physical basis of Laplace’s equation, the wave equation and the diffusion equation.
General method of separation of variables in Cartesian, cylindrical and spherical
co ordinates. Legendre’s equation: derivation, solutions including explicit forms of
P
0
,
P
1
and
P
2
, orthogonality. Bessel’s equation of integer order as an example of a
self-adjoint eigenvalue problem with non-trivial weight.
Examples including potentials on rectangular and circular domains and on a spherical
domain (axisymmetric case only), waves on a finite string and heat flow down a
semi-infinite rod. [5]
Inhomogeneous ODEs: Green’s functions
Properties of the Dirac delta function. Initial value problems and forced problems with
two fixed end points; solution using Green’s functions. Eigenfunction expansions of the
delta f unction and Green’s functions. [4]
Fourier transforms
Fourier transforms: definition and simple properties; inversion and convolution theorems.
The discrete Fourier transform. Examples of application to linear systems. Relationship
of transfer function to Green’s function for initial value problems. [4]
PDEs on unbounded domains
Classification of PDEs in two independent variables. Well posedness. Solution by
the method of characteristics. Green’s functions for PDEs in 1, 2 and 3 independent
variables; fundamental solutions of the wave equation, Laplace’s equation and the
diffusion equation. The method of images. Application to the forced wave equation,
Poisson’s equation and forced diffusion equation. Transient solutions of diffusion
problems: the error function. [6]
Contents
0 Introduction
1 Vector spaces
2 Fourier series
2.1 Fourier series
2.2 Convergence of Fourier series
2.3 Differentiability and Fourier series
3 Sturm-Liouville Theory
3.1 Sturm-Liouville operators
3.2 Least squares approximation
4 Partial differential equations
4.1 Laplace’s equation
4.2 Laplace’s equation in the unit disk in R
2
4.3 Separation of variables
4.4 Laplace’s equation in spherical polar coordinates
4.4.1 Laplace’s equation in spherical polar coordinates
4.4.2 Legendre Polynomials
4.4.3 Solution to radial part
4.5 Multipole expansions for Laplace’s equation
4.6 Laplace’s equation in cylindrical coordinates
4.7 The heat equation
4.8 The wave equation
5 Distributions
5.1 Distributions
5.2 Green’s functions
5.3 Green’s functions for initial value problems
6 Fourier transforms
6.1 The Fourier transform
6.2 The Fourier inversion theorem
6.3 Parseval’s theorem for Fourier transforms
6.4 A word of warning
6.5 Fourier transformation of distributions
6.6 Linear systems and response functions
6.7 General form of transfer function
6.8 The discrete Fourier transform
6.9 The fast Fourier transform*
7 More partial differential equations
7.1 Well-posedness
7.2 Method of characteristics
7.3 Characteristics for 2nd order partial differential equations
7.4 Green’s functions for PDEs on R
n
7.5 Poisson’s equation
0 Introduction
In the previous courses, the (partial) differential equations we have seen are
mostly linear. For example, we have Laplace’s equation:
2
φ
x
2
+
φ
y
2
= 0,
and the heat equation:
φ
t
= κ
2
φ
x
2
+
2
φ
y
2
.
The Schr¨odinger’ equation in quantum mechanics is also linear:
i~
Φ
t
=
~
2
2m
2
φ
x
2
+ V (x)Φ(x).
By being linear, these equations have the property that if
φ
1
, φ
2
are solutions,
then so are λ
1
φ
1
+ λ
2
φ
2
(for any constants λ
i
).
Why are all these linear? In general, if we just randomly write down a
differential equation, most likely it is not going to be linear. So where did the
linearity of equations of physics come from?
The answer is that the real world is not linear in general. However, often we
are not looking for a completely accurate and precise description of the universe.
When we have low energy/speed/whatever, we can often quite accurately approx-
imate reality by a linear equation. For example, the equation of general relativity
is very complicated and nowhere near being linear, but for small masses and
velocities, they reduce to Newton’s law of gravitation, which is linear.
The only exception to this seems to be Schr¨odinger’s equation. While there
are many theories and equations that superseded the Schr¨odinger equation, these
are all still linear in nature. It seems that linearity is the thing that underpins
quantum mechanics.
Due to the prevalence of linear equations, it is rather important that we
understand these equations well, and this is our primary objective of the course.
1 Vector spaces
When dealing with functions and differential equations, we will often think of
the space of functions as a vector space. In many cases, we will try to find a
“basis” for our space of functions, and expand our functions in terms of the basis.
Under different situations, we would want to use a different basis for our space.
For example, when dealing with periodic functions, we will want to pick basis
elements that are themselves periodic. In other situations, these basis elements
would not be that helpful.
A familiar example would be the Taylor series, where we try to approximate
a function f by
f(x) =
X
n=0
f
(n)
(0)
n!
x
n
.
Here we are thinking of
{x
n
:
n N}
as the basis of our space, and trying to
approximate an arbitrary function as a sum of the basis elements. When writing
the function
f
as a sum like this, it is of course important to consider whether
the sum converges, and when it does, whether it actually converges back to f.
Another issue of concern is if we have a general set of basis functions
{y
n
}
,
how can we find the coefficients
c
n
such that
f
(
x
) =
P
c
n
y
n
(
x
)? This is the
bit where linear algebra comes in. Finding these coefficients is something we
understand well in linear algebra, and we will attempt to borrow the results and
apply them to our space of functions.
Another concept we would want to borrow is eigenvalues and eigenfunctions,
as well as self-adjoint (“Hermitian”) operators. As we go along the course, we
will see some close connections between functions and vector spaces, and we can
often get inspirations from linear algebra.
Of course, there is no guarantee that the results from linear algebra would
apply directly, since most of our linear algebra results was about finite basis and
finite linear sums. However, it is often a good starting point, and usually works
when dealing with sufficiently nice functions.
We start with some preliminary definitions, which should be familiar from
IA Vectors and Matrices and/or IB Linear Algebra.
Definition
(Vector space)
.
A vector space over
C
(or
R
) is a set
V
with an
operation + which obeys
(i) u + v = v + u (commutativity)
(ii) (u + v) + w = u + (v + w) (associativity)
(iii) There is some 0 V such that 0 + u = u for all u (identity)
We can also multiply vectors by a scalars λ C, which satisfies
(i) λ(µv) = (λµ)v (associativity)
(ii) λ(u + v) = λu + λv (distributivity in V )
(iii) (λ + µ)u = λu + λv (distributivity in C)
(iv) 1v = v (identity)
Often, we wouldn’t have just a vector space. We usually give them some
additional structure, such as an inner product.
Definition
(Inner product)
.
An inner product on
V
is a map (
·, ·
) :
V ×V C
that satisfies
(i) (u, λv) = λ(u, v) (linearity in second argument)
(ii) (u, v + w) = (u, v) + (u, w) (additivity)
(iii) (u, v) = (v, u)
(conjugate symmetry)
(iv) (u, u) 0, with equality iff u = 0 (positivity)
Note that the positivity condition makes sense since conjugate symmetry entails
that (u, u) R.
The inner product in turn defines a norm
kuk
=
p
(u, u)
that provides the
notion of length and distance.
It is important to note that we only have linearity in the second argument.
For the first argument, we have (λu, v) = (v, λu)
= λ
(v, u)
= λ
(u, v).
Definition
(Basis)
.
A set of vectors
{v
1
, v
2
, ··· , v
n
}
form a basis of
V
iff any
u V can be uniquely written as a linear combination
u =
n
X
i=1
λ
i
v
i
for some scalars
λ
i
. The dimension of a vector space is the number of basis
vectors in its basis.
A basis is orthogonal (with respect to the inner product) if (
v
i
, v
j
) = 0
whenever i 6= j.
A basis is orthonormal (with respect to the inner product) if it is orthogonal
and (v
i
, v
i
) = 1 for all i.
Orthonormal bases are the nice bases, and these are what we want to work
with.
Given an orthonormal basis, we can use the inner product to find the
expansion of any u V in terms of the basis, for if
u =
X
i
λ
i
v
i
,
taking the inner product with v
j
gives
(v
j
, u) =
v
j
,
X
i
λ
i
v
i
!
=
X
i
λ
i
(v
j
, v
i
) = λ
j
,
using additivity and linearity. Hence we get the general formula
λ
i
= (v
i
, u).
We have seen all these so far in IA Vectors and Matrices, where a vector is a list
of finitely many numbers. However, functions can also be thought of as elements
of an (infinite dimensional) vector space.
Suppose we have
f, g
:
C
. Then we can define the sum
f
+
g
by
(f + g)(x) = f(x) + g(x). Given scalar λ, we can also define (λf )(x) = λf (x).
This also makes intuitive sense. We can simply view a functions as a list
of numbers, where we list out the values of
f
at each point. The list could be
infinite, but a list nonetheless.
Most of the time, we don’t want to look at the set of all functions. That
would be too huge and uninteresting. A natural class of functions to consider
would be the set of solutions to some particular differential solution. However,
this doesn’t always work. For this class to actually be a vector space, the sum
of two solutions and the scalar multiple of a solution must also be a solution.
This is exactly the requirement that the differential equation is linear. Hence,
the set of solutions to a linear differential equation would form a vector space.
Linearity pops up again.
Now what about the inner product? A natural definition is
(f, g) =
Z
Σ
f(x)
g(x) dµ,
where
µ
is some measure. For example, we could integrate d
x
, or d
x
2
. This
measure specifies how much weighting we give to each point x.
Why does this definition make sense? Recall that the usual inner product
on finite-dimensional vector spaces is
P
v
i
w
i
. Here we are just summing the
different components of
v
and
w
. We just said we can think of the function
f
as
a list of all its values, and this integral is just the sum of all components of
f
and g.
Example. Let Σ = [a, b]. Then we could take
(f, g) =
Z
b
a
f(x)
g(x) dx.
Alternatively, let Σ = D
2
R
2
be the unit disk. Then we could have
(f, g) =
Z
1
0
Z
2π
0
f(r, θ)
g(r, θ) dθ r dr
Note that we were careful and said that d
µ
is “some measure”. Here we are
integrating against d
θ r
d
r
. We will later see cases where this can be even more
complicated.
If Σ has a boundary, we will often want to restrict our functions to take
particular values on the boundary, known as boundary conditions. Often, we
want the boundary conditions to preserve linearity. We call these nice boundary
conditions homogeneous conditions.
Definition
(Homogeneous boundary conditions)
.
A boundary condition is
homogeneous if whenever
f
and
g
satisfy the boundary conditions, then so does
λf + µg for any λ, µ C (or R).
Example.
Let Σ = [
a, b
]. We could require that
f
(
a
) + 7
f
0
(
b
) = 0, or maybe
f
(
a
) + 3
f
00
(
a
) = 0. These are examples of homogeneous boundary conditions.
On the other hand, the requirement f(a) = 1 is not homogeneous.
2 Fourier series
The first type of functions we will consider is periodic functions.
Definition
(Periodic function)
.
A function
f
is periodic if there is some fixed
R such that f(x + R) = f (x) for all x.
However, it is often much more convenient to think of this as a function
f
:
S
1
C
from unit circle to
C
, and parametrize our function by an angle
θ
ranging from 0 to 2π.
Now why do we care about periodic functions? Apart from the fact that
genuine periodic functions exist, we can also use them to model functions on a
compact domain. For example, if we have a function defined on [0
,
1], we can
pretend it is a periodic function on
R
by making infinitely many copies of the
function to the intervals [1, 2], [2, 3] etc.
x x
2.1 Fourier series
So. As mentioned in the previous chapter, we want to find a set of “basis
functions” for periodic functions. We could go with the simplest case of periodic
functions we know of the exponential function
e
inθ
. These functions have a
period of 2
π
, and are rather easy to work with. We all know how to integrate
and differentiate the exponential function.
More importantly, this set of basis functions is orthogonal. We have
(e
imθ
, e
inθ
) =
Z
π
π
e
imθ
e
inθ
dθ =
Z
π
π
e
i(nm)θ
dθ =
(
2π n = m
0 n 6= m
= 2πδ
nm
We can normalize these to get a set of orthonormal functions
n
1
2π
e
inθ
: n Z
o
.
Fourier’s idea was to use this as a basis for any periodic function. Fourier
claimed that any f : S
1
C can be expanded in this basis:
f(θ) =
X
nZ
ˆ
f
n
e
inθ
,
where
ˆ
f
n
=
1
2π
(e
inθ
, f) =
1
2π
Z
π
π
e
inθ
f(θ) dθ.
These really should be defined as
f
(
θ
) =
P
ˆ
f
n
e
inθ
2π
with
ˆ
f
n
=
e
inθ
2π
, f
, but for
convenience reasons, we move all the constant factors to the
ˆ
f
n
coefficients.
We can consider the special case where
f
:
S
1
R
is a real function. We
might want to make our expansion look a bit more “real”. We get
(
ˆ
f
n
)
=
1
2π
Z
π
π
e
inθ
f(θ) dθ
=
1
2π
Z
π
π
e
inθ
f(θ) dθ =
ˆ
f
n
.
So we can replace our Fourier series by
f(θ) =
ˆ
f
0
+
X
n=1
ˆ
f
n
e
inθ
+
ˆ
f
n
e
inθ
=
ˆ
f
0
+
X
n=1
ˆ
f
n
e
inθ
+
ˆ
f
n
e
inθ
.
Setting
ˆ
f
n
=
a
n
+ ib
n
2
, we can write this as
f(θ) =
ˆ
f
0
+
X
n=1
(a
n
cos + b
n
sin )
=
a
0
2
+
X
n=1
(a
n
cos + b
n
sin ).
Here the coefficients are
a
n
=
1
π
Z
π
π
cos f(θ) dθ, b
n
=
1
π
Z
π
π
sin f(θ) dθ.
This is an alternative formulation of the Fourier series in terms of sin and cos.
So when given a real function, which expansion should we use? It depends. If
our function is odd (or even), it would be useful to pick the sine/cosine expansion,
since the cosine (or sine) terms will simply disappear. On the other hand, if we
want to stick our function into a differential equation, exponential functions are
usually more helpful.
2.2 Convergence of Fourier series
When Fourier first proposed the idea of a Fourier series, people didn’t really
believe in him. How can we be sure that the infinite series actually converges?
It turns out that in many cases, they don’t.
To investigate the convergence of the series, we define the partial Fourier
sum as
S
n
f =
n
X
m=n
ˆ
f
m
e
imθ
.
The question we want to answer is whether
S
n
f
“converges” to
f
. Here we
have to be careful with what we mean by convergence. As we (might) have seen
in Analysis, there are many ways of defining convergence of functions. If we
have a “norm” on the space of functions, we can define convergence to mean
lim
n→∞
kS
n
f fk = 0. Our “norm” can be defined as
kS
n
f fk =
1
2π
Z
π
π
|S
n
f(θ) f(θ)|
2
dθ.
However, this doesn’t really work if
S
n
f
and
f
can be arbitrary functions, as in
this does not necessarily have to be a norm. Indeed, if
lim
n→∞
S
n
f
differs from
f
on finitely or countably many points, the integral will still be zero. In particular,
lim S
n
f
and
f
can differ at all rational points, but this definition will say that
S
n
f converges to f.
Hence another possible definition of convergence is to require
lim
n→∞
S
n
f(θ) f(θ) = 0
for all
θ
. This is known as pointwise convergence. However, this is often too
weak a notion. We can ask for more, and require that the rate of convergent is
independent of θ. This is known as uniform convergence, and is defined by
lim
n→∞
sup
θ
|S
n
f(θ) f(θ)| = 0.
Of course, with different definitions of convergence, we can get different answers
to whether it converges. Unfortunately, even if we manage to get our definition
of convergence right, it is still difficult to come up with a criterion to decide if a
function has a convergent Fourier Series.
It is not difficult to prove a general criterion for convergence in norm, but
we will not do that, since that is analysis. If interested, one should take the IID
Analysis of Functions course. In this course, instead of trying to come up with
something general, let’s look at an example instead.
Example. Consider the sawtooth function f(θ) = θ.
3π
π
π
3π
Note that this function is discontinuous at odd multiples of π.
The Fourier coefficients (for n 6= 0) are
ˆ
f
n
=
1
2π
Z
π
π
e
inθ
θ dθ =
1
2πin
e
inθ
θ
π
π
+
1
2πin
Z
π
π
e
inθ
dθ =
(1)
n+1
in
.
We also have
ˆ
f
0
= 0.
Hence we have
θ =
X
n6=0
(1)
n+1
in
e
inθ
.
It turns out that this series converges to the sawtooth for all
θ 6
= (2
m
+ 1)
π
, i.e.
everywhere that the sawtooth is continuous.
Let’s look explicitly at the case where
θ
=
π
. Each term of the partial Fourier
series is zero. So we can say that the Fourier series converges to 0. This is the
average value of lim
ε0
f(π ± ε).
This is typical. At an isolated discontinuity, the Fourier series is the average
of the limiting values of the original function as we approach from either side.
2.3 Differentiability and Fourier series
Integration is a smoothening operator. If we take, say, the step function
Θ(x) =
(
1 x > 0
0 x < 0
.
Then the integral is given by
Z
Θ(x) dx =
(
x x > 0
0 x < 0
.
This is now a continuous function. If we integrate it again, we make the positive
side look like a quadratic curve, and it is now differentiable.
x
y
R
dx
x
y
On the other hand, when we differentiate a function, we generally make it worse.
For example, when we differentiate the continuous function
R
Θ(
x
) d
x
, we obtain
the discontinuous function Θ(
x
). If we attempt to differentiate this again, we
get the δ-(non)-function.
Hence, it is often helpful to characterize the “smoothness” of a function by
how many times we can differentiate it. It turns out this is rather relevant to
the behaviour of the Fourier series.
Suppose that we have a function that is itself continuous and whose first
m
1 derivatives are also continuous, but
f
(m)
has isolated discontinuities at
{θ
1
, θ
2
, θ
3
, ··· , θ
r
}.
We can look at the kth Fourier coefficient:
ˆ
f
k
=
1
2π
Z
π
π
e
ikθ
f(θ) dθ
=
1
2πik
e
ikθ
f(θ)
π
π
+
1
2πik
Z
π
π
e
ikθ
f
0
(θ) dθ
The first term vanishes since
f
(
θ
) is continuous and takes the same value at
π
and π. So we are left with
=
1
2πik
Z
π
π
e
ikθ
f
0
(θ) dθ
= ···
=
1
(ik)
m
1
2π
Z
π
π
e
ikθ
f
(m)
(θ) dθ.
Now we have to be careful, since
f
(m)
is no longer continuous. However provided
that
f
(m)
is everywhere finite, we can approximate this by removing small strips
(
θ
i
ε, θ
i
+
ε
) from our domain of integration, and take the limit
ε
0. We can
write this as
= lim
ε0
1
(ik)
m
1
2π
Z
θ
1
ε
π
+
Z
θ
2
ε
θ
1
+ε
+ ··· +
Z
π
θ
r
+ε
e
ikθ
f
(m)
(θ) dθ
=
1
(ik)
m+1
1
2π
"
r
X
s=1
(f
(m)
(θ
+
s
) f
(m)
(θ
s
)) +
Z
(π)\θ
e
ikθ
f
(m+1)
(θ) dθ
#
.
We now have to stop. So
ˆ
f
k
decays like
1
k
m+1
if our function and its (
m
1)
derivatives are continuous. This means that if a function is more differentiable,
then the coefficients decay more quickly.
This makes sense. If we have a rather smooth function, then we would expect
the first few Fourier terms (with low frequency) to account for most of the
variation of f. Hence the coefficients decay really quickly.
However, if the function is jiggly and bumps around all the time, we would
expect to need some higher frequency terms to account for the minute variation.
Hence the terms would not decay away that quickly. So in general, if we can
differentiate it more times, then the terms should decay quicker.
An important result, which is in some sense a Linear Algebra result, is
Parseval’s theorem.
Theorem (Parseval’s theorem).
(f, f) =
Z
π
π
|f(θ)|
2
dθ = 2π
X
nZ
|
ˆ
f
n
|
2
Proof.
(f, f) =
Z
π
π
|f(θ)|
2
dθ
=
Z
π
π
X
mZ
ˆ
f
m
e
imθ
!
X
nZ
ˆ
f
n
e
inθ
!
dθ
=
X
m,nZ
ˆ
f
m
ˆ
f
n
Z
π
π
e
i(nm)θ
dθ
= 2π
X
m,nZ
ˆ
f
m
ˆ
f
n
δ
mn
= 2π
X
nZ
|
ˆ
f
n
|
2
Note that this is not a fully rigorous proof, since we assumed not only that
the Fourier series converges to the function, but also that we could commute the
infinite sums. However, for the purposes of an applied course, this is sufficient.
Last time, we computed that the sawtooth function
f
(
θ
) =
θ
has Fourier
coefficients
ˆ
f
0
= 0,
ˆ
f
n
=
i(1)
n
n
for n 6= 0.
But why do we care? It turns out this has some applications in number theory.
You might have heard of the Riemann ζ-function, defined by
ζ(s) =
X
n=1
1
n
s
.
We will show that this obeys the property that for any
m
,
ζ
(2
m
) =
π
2m
q
for
some
q Q
. This may not be obvious at first sight. So let’s apply Parseval’s
theorem for the sawtooth defined by
f
(
θ
) =
θ
. By direct computation, we know
that
(f, f) =
Z
π
π
θ
2
dθ =
2π
3
3
.
However, by Parseval’s theorem, we know that
(f, f) = 2π
X
nZ
|
ˆ
f
n
|
2
= 4π
X
n=1
1
n
2
.
Putting these together, we learn that
X
n=1
1
n
2
= ζ(2) =
π
2
6
.
We have just done it for the case where
m
= 1. But if we integrate the sawtooth
function repeatedly, then we can get the general result for all m.
At this point, we might ask, why are we choosing these
e
imθ
as our basis?
Surely there are a lot of different sets of basis we can use. For example, in finite
dimensions, we can just arbitrary choose random vectors (that are not linearly
dependent) to get a set of basis vectors. However, in practice, we don’t pick
them randomly. We often choose a basis that can reveal the symmetry of a
system. For example, if we have a rotationally symmetric system, we would like
to use polar coordinates. Similarly, if we have periodic functions, then
e
imθ
is
often a good choice of basis.
3 Sturm-Liouville Theory
3.1 Sturm-Liouville operators
In finite dimensions, we often consider linear maps
M
:
V W
. If
{v
i
}
is a
basis for
V
and
{w
i
}
is a basis for
W
, then we can represent the map by a
matrix with entries
M
ai
= (w
a
, Mv
i
).
A map
M
:
V V
is called self-adjoint if
M
=
M
as matrices. However, it is
not obvious how we can extend this notion to arbitrary maps between arbitrary
vector spaces (with an inner product) when they cannot be represented by a
matrix.
Instead, we make the following definitions:
Definition
(Adjoint and self-adjoint)
.
The adjoint
B
of a map
A
:
V V
is a
map such that
(Bu, v) = (u, Av)
for all vectors u, v V . A map is then self-adjoint if
(Mu, v) = (u, M v).
Self-adjoint matrices come with a natural basis. Recall that the eigenvalues
of a matrix are the roots of
det
(
M λI
) = 0. The eigenvector corresponding to
an eigenvalue λ is defined by Mv
i
= λ
i
v
i
.
In general, eigenvalues can be any complex number. However, self-adjoint
maps have real eigenvalues. Suppose
Mv
i
= λ
i
v
i
.
Then we have
λ
i
(v
i
, v
i
) = (v
i
, Mv
i
) = (Mv
i
, v
i
) = λ
i
(v
i
, v
i
).
So λ
i
= λ
i
.
Furthermore, eigenvectors with distinct eigenvalues are orthogonal with
respect to the inner product. Suppose that
Mv
i
= λ
i
v
i
, M v
j
= λ
j
v
j
.
Then
λ
i
(v
j
, v
i
) = (v
j
, Mv
i
) = (Mv
j
, v
i
) = λ
j
(v
j
, v
i
).
Since λ
i
6= λ
j
, we must have (v
j
, v
i
) = 0.
Knowing eigenvalues and eigenvalues gives a neat way so solve linear equations
of the form
Mu = f.
Here we are given
M
and
f
, and want to find
u
. Of course, the answer is
u = M
1
f. However, if we expand in terms of eigenvectors, we obtain
Mu = M
X
u
i
v
i
=
X
u
i
λ
i
v
i
.
Hence we have
X
u
i
λ
i
v
i
=
X
f
i
v
i
.
Taking the inner product with v
j
, we know that
u
j
=
f
j
λ
j
.
So far, these are all things from IA Vectors and Matrices. Sturm-Liouville theory
is the infinite-dimensional analogue.
In our vector space of differentiable functions, our “matrices” would be linear
differential operators L. For example, we could have
L = A
p
(x)
d
p
dx
p
+ A
p1
(x)
d
p1
dx
p1
+ ··· + A
1
(x)
d
dx
+ A
0
(x).
It is an easy check that this is in fact linear.
We say L has order p if the highest derivative that appears is
d
p
dx
p
.
In most applications, we will be interested in the case
p
= 2. When will our
L be self-adjoint?
In the p = 2 case, we have
Ly = P
d
2
y
dx
2
+ R
dy
dx
Qy
= P
d
2
y
dx
2
+
R
P
dy
dx
Q
P
y
= P
e
R
R
P
dx
d
dx
e
R
R
P
dx
dy
dx
Q
P
y
Let p = exp
R
R
P
dx
. Then we can write this as
= P p
1
d
dx
p
dy
dx
Q
P
py
.
We further define
q
=
Q
P
p
. We also drop a factor of
P p
1
. Then we are left with
L =
d
dx
p(x)
d
dx
q(x).
This is the Sturm-Liouville form of the operator. Now let’s compute (
f, Lg
). We
integrate by parts numerous times to obtain
(f, Lg) =
Z
b
a
f
d
dx
p
dg
dx
qg
dx
= [f
pg
0
]
b
a
Z
b
a
df
dx
p
dg
dx
+ f
qg
dx
= [f
pg
0
f
0∗
pg]
b
a
+
Z
b
a
d
dx
p
df
dx
qf
g dx
= [(f
g
0
f
0∗
g)p]
b
a
+ (Lf, g),
assuming that p, q are real.
So 2nd order linear differential operators are self-adjoint with respect to this
norm if
p, q
are real and the boundary terms vanish. When do the boundary
terms vanish? One possibility is when
p
is periodic (with the right period), or if
we constrain f and g to be periodic.
Example. We can consider a simple case, where
L =
d
2
dx
2
.
Here we have p = 1, q = 0. If we ask for functions to be periodic on [a, b], then
Z
b
a
f
d
2
g
dx
2
dx =
Z
b
a
d
2
f
dx
2
g dx.
Note that it is important that we have a second-order differential operator. If it
is first-order, then we would have a negative sign, since we integrated by parts
once.
Just as in finite dimensions, self-adjoint operators have eigenfunctions and
eigenvalues with special properties. First, we define a more sophisticated inner
product.
Definition
(Inner product with weight)
.
An inner product with weight
w
,
written ( ·, ·)
w
, is defined by
(f, g)
w
=
Z
b
a
f
(x)g(x)w(x) dx,
where w is real, non-negative, and has only finitely many zeroes.
Why do we want a weight
w
(
x
)? In the future, we might want to work with
the unit disk, instead of a square in
R
2
. When we want to use polar coordinates,
we will have to integrate with
r
d
r
d
θ
, instead of just d
r
d
θ
. Hence we need the
weight of
r
. Also, we allow it to have finitely many zeroes, so that the radius
can be 0 at the origin.
Why can’t we have more zeroes? We want the inner product to keep the
property that (
f, f
)
w
= 0 iff
f
= 0 (for continuous
f
). If
w
is zero at too many
places, then the inner product could be zero without f being zero.
We now define what it means to be an eigenfunction.
Definition
(Eigenfunction with weight)
.
An eigenfunction with weight
w
of
L
is a function y : [a, b] C obeying the differential equation
Ly = λwy,
where λ C is the eigenvalue.
This might be strange at first sight. It seems like we can take any nonsense
y
, apply
L
, to get some nonsense
Ly
. But then it is fine, since we can write it
as some nonsense
w
times our original
y
. So any function is an eigenfunction?
No! There are many constraints
w
has to satisfy, like being positive, real and
having finitely many zeroes. It turns out this severely restraints what values
y
can take, so not everything will be an eigenfunction. In fact we can develop
this theory without the weight function
w
. However, weight functions are much
more convenient when, say, dealing with the unit disk.
Proposition. The eigenvalues of a Sturm-Liouville operator are real.
Proof. Suppose Ly
i
= λ
i
wy
i
. Then
λ
i
(y
i
, y
i
)
w
= λ
i
(y
i
, wy
i
) = (y
i
, Ly
i
) = (Ly
i
, y
i
) = (λ
i
wy
i
, y
i
) = λ
i
(y
i
, y
i
)
w
.
Since (y
i
, y
i
)
w
6= 0, we have λ
i
= λ
i
.
Note that the first and last terms use the weighted inner product, but the
middle terms use the unweighted inner product.
Proposition.
Eigenfunctions with different eigenvalues (but same weight) are
orthogonal.
Proof. Let Ly
i
= λ
i
wy
i
and Ly
j
= λ
j
wy
j
. Then
λ
i
(y
j
, y
i
)
w
= (y
j
, Ly
i
) = (Ly
j
, y
i
) = λ
j
(y
j
, y
i
)
w
.
Since λ
i
6= λ
j
, we must have (y
j
, y
i
)
w
= 0.
Those were pretty straightforward manipulations. However, the main results
of Sturm–Liouville theory are significantly harder, and we will not prove them.
We shall just state them and explore some examples.
Theorem. On a compact domain, the eigenvalues λ
1
, λ
2
, ··· form a countably
infinite sequence and are discrete.
This will be a rather helpful result in quantum mechanics, since in quantum
mechanics, the possible values of, say, the energy are the eigenvalues of the
Hamiltonian operator. Then this result says that the possible values of the
energy are discrete and form an infinite sequence.
Note here the word compact. In quantum mechanics, if we restrict a particle
in a well [0
,
1], then it will have quantized energy level since the domain is
compact. However, if the particle is free, then it can have any energy at all since
we no longer have a compact domain. Similarly, angular momentum is quantized,
since it describe rotations, which takes values in S
1
, which is compact.
Theorem.
The eigenfunctions are complete: any function
f
: [
a, b
]
C
(obeying
appropriate boundary conditions) can be expanded as
f(x) =
X
n
ˆ
f
n
y
n
(x),
where
ˆ
f
n
=
Z
y
n
(x)f(x)w(x) dx.
Example.
Let [
a, b
] = [
L, L
],
L
=
d
2
dx
2
,
w
= 1, restricting to periodic functions
Then our eigenfunction obeys
d
2
y
n
dx
2
= λ
n
y
n
(x),
Then our eigenfunctions are
y
n
(x) = exp
inπx
L
with eigenvalues
λ
n
=
n
2
π
2
L
2
for n Z. This is just the Fourier series!
Example
(Hermite polynomials)
.
We are going to cheat a little bit and pick
our domain to be R. We want to study the differential equation
1
2
H
00
xH
0
= λH,
with H : R C. We want to put this in Sturm-Liouville form. We have
p(x) = exp
Z
x
0
2t dt
= e
x
2
,
ignoring constant factors. Then q(x) = 0. We can rewrite this as
d
dx
e
x
2
dH
dx
= 2λe
x
2
H(x).
So we take our weight function to be w(x) = e
x
2
.
We now ask that
H
(
x
) grows at most polynomially as
|x|
. In particular,
we want
e
x
2
H
(
x
)
2
0. This ensures that the boundary terms from integration
by parts vanish at the infinite boundary, so that our Sturm-Liouville operator is
self-adjoint.
The eigenfunctions turn out to be
H
n
(x) = (1)
n
e
x
2
d
n
dx
n
e
x
2
.
These are known as the Hermite polynomials. Note that these are indeed
polynomials. When we differentiate the
e
x
2
term many times, we get a lot
of things from the product rule, but they will always keep an
e
x
2
, which will
ultimately cancel with e
x
2
.
Just as for matrices, we can use the eigenfunction expansion to solve forced
differential equations. For example, if might want to solve
Lg = f(x),
where f (x) is a forcing term. We can write this as
Lg = w(x)F (x).
We expand our g as
g(x) =
X
nZ
ˆg
n
y
n
(x).
Then by linearity,
Lg =
X
nZ
ˆg
n
Ly
n
(x) =
X
nZ
ˆg
n
λ
n
w(x)y
n
(x).
We can also expand our forcing function as
w(x)F(x) = w(x)
X
nZ
ˆ
F
n
y
n
(x)
!
.
Taking the (regular) inner product with
y
m
(
x
) (and noting orthogonality of
eigenfunctions), we obtain
w(xg
m
λ
m
= w(x)
ˆ
F
m
This tells us that
ˆg
m
=
ˆ
F
m
λ
m
.
So we have
g(x) =
X
nZ
ˆ
F
n
λ
n
y
n
(x),
provided all λ
n
are non-zero.
This is a systematic way of solving forced differential equations. We used to
solve these by “being smart”. We just looked at the forcing term and tried to
guess what would work. Unsurprisingly, this approach does not succeed all the
time. Thus it is helpful to have a systematic way of solving the equations.
It is often helpful to rewrite this into another form, using the fact that
ˆ
F
n
= (y
n
, F )
w
. So we have
g(x) =
X
nZ
1
λ
n
(y
n
, F )
w
y
n
(x) =
Z
b
a
X
nZ
1
λ
n
y
n
(t)y
n
(x)w(t)F(t) dt.
Note that we swapped the sum and the integral, which is in general a dangerous
thing to do, but we don’t really care because this is an applied course. We can
further write the above as
g(x) =
Z
b
a
G(x, t)F (t)w(t) dt,
where G(x, t) is the infinite sum
G(x, t) =
X
nZ
1
λ
n
y
n
(t)y
n
(x).
We call this the Green’s function. Note that this depends on
λ
n
and
y
n
only.
It depends on the differential operator
L
, but not the forcing term
F
. We can
think of this as something like the “inverse matrix”, which we can use to solve
the forced differential equation for any forcing term.
Recall that for a matrix, the inverse exists if the determinant is non-zero,
which is true if the eigenvalues are all non-zero. Similarly, here a necessary
condition for the Green’s function to exist is that all the eigenvalues are non-zero.
We now have a second version of Parseval’s theorem.
Theorem (Parseval’s theorem II).
(f, f)
w
=
X
nZ
|
ˆ
f
n
|
2
.
Proof. We have
(f, f)
w
=
Z
f
(x)f(x)w(x) dx
=
X
n,mZ
Z
ˆ
f
n
y
n
(x)
ˆ
f
m
y
m
(x)w(x) dx
=
X
n,mZ
ˆ
f
n
ˆ
f
m
(y
n
, y
m
)
w
=
X
nZ
|
ˆ
f
n
|
2
.
3.2 Least squares approximation
So far, we have expanded functions in terms of infinite series. However, in real
life, when we ask a computer to do this for us, it is incapable of storing and
calculating an infinite series. So we need to truncate it at some point.
Suppose we approximate some function
f
: Ω
C
by a finite set of eigen-
functions y
n
(x). Suppose we write the approximation g as
g(x) =
n
X
k=1
c
k
y
k
(x).
The objective here is to figure out what values of the coefficients
c
k
are “the
best”, i.e. can make g represent f as closely as possible.
Firstly, we have to make it explicit what we mean by “as closely as possible”.
Here we take this to mean that we want to minimize the
w
-norm (
f g, f g
)
w
.
By linearity of the norm, we know that
(f g, f g)
w
= (f, f)
w
+ (g, g)
w
(f, g)
w
(g, f)
w
.
To minimize this norm, we want
0 =
c
j
(f g, f g)
w
=
c
j
[(f, f)
w
+ (g, g)
w
(f, g)
w
(g, f)
w
].
We know that the (
f, f
)
w
term vanishes since it does not depend on
c
k
. Expanding
our definitions of g, we can get
0 =
c
j
X
i=1
|c
k
|
n
n
X
k=1
ˆ
f
k
c
k
n
X
k=1
c
k
ˆ
f
k
!
= c
j
ˆ
f
j
.
Note that here we are treating
c
j
and
c
j
as distinct quantities. So when we vary
c
j
,
c
j
is unchanged. To formally justify this treatment, we can vary the real and
imaginary parts separately.
Hence, the extremum is achieved at
c
j
=
ˆ
f
j
. Similarly, varying with respect
to c
j
, we get that c
j
=
ˆ
f
j
.
To check that this is indeed an minimum, we can look at the second-derivatives
2
c
i
c
j
(f g, f g)
w
=
2
c
i
c
j
(f g, f g)
w
= 0,
while
2
c
i
c
j
(f g, f g)
w
= δ
ij
0.
Hence this is indeed a minimum.
Thus we know that (f g, f g)
w
is minimized over all g(x) when
c
k
=
ˆ
f
k
= (y
k
, f)
w
.
These are exactly the coefficients in our infinite expansion. Hence if we truncate
our series at an arbitrary point, it is still the best approximation we can get.
4 Partial differential equations
So far, we have come up with a lot of general theory about functions and stuff.
We are going to look into some applications of these. In this section, we will
develop the general technique of separation of variables, and apply this to many
many problems.
4.1 Laplace’s equation
Definition
(Laplace’s equation)
.
Laplace’s equation on
R
n
says that a (twice-
differentiable) equation φ : R
n
C obeys
2
φ = 0,
where
2
=
n
X
i=1
2
x
2
i
.
This equation arises in many situations. For example, if we have a conservative
force
F
=
−∇φ
, and we are in a region where the force obeys the source-free
Gauss’ law
·F
= 0 (e.g. when there is no electric charge in an electromagnetic
field), then we get Laplace’s equation.
It also arises as a special case of the heat equation
φ
t
=
κ
2
φ
, and wave
equation
2
φ
t
2
= c
2
2
φ, when there is no time dependence.
Definition
(Harmonic functions)
.
Functions that obey Laplace’s equation are
often called harmonic functions.
Proposition.
Let be a compact domain, and
be its boundary. Let
f : R. Then there is a unique solution to
2
φ = 0 on with φ|
= f.
Proof.
Suppose
φ
1
and
φ
2
are both solutions such that
φ
1
|
=
φ
2
|
=
f
. Then
let Φ = φ
1
φ
2
. So Φ = 0 on the boundary. So we have
0 =
Z
Φ
2
Φ d
n
x =
Z
(Φ) · (Φ) dx +
Z
ΦΦ · n d
n1
x.
We note that the second term vanishes since Φ = 0 on the boundary. So we have
0 =
Z
(Φ) · (Φ) dx.
However, we know that (
Φ)
·
(
Φ)
0 with equality iff
Φ = 0. Hence
Φ is constant throughout Ω. Since at the boundary, Φ = 0, we have Φ = 0
throughout, i.e. φ
1
= φ
2
.
Asking that
φ|
=
f
(
x
) is a Dirichlet boundary condition. We can instead
ask
n ·φ|
=
g
, a Neumann condition. In this case, we have a unique solution
up to a constant factor.
These we’ve all seen in IA Vector Calculus, but this time in arbitrary dimen-
sions.
4.2 Laplace’s equation in the unit disk in R
2
Let Ω = {(x, y) R
2
: x
2
+ y
2
1}. Then Laplace’s equation becomes
0 =
2
φ =
2
φ
x
2
+
2
φ
y
2
=
2
φ
z ¯z
,
where we define z = x + iy, ¯z = x iy. Then the general solution is
φ(z, ¯z) = ψ(z) + χ(¯z)
for some ψ, χ.
Suppose we want a solution obeying the Dirichlet boundary condition
φ
(
z, ¯z
)
|
=
f
(
θ
), where the boundary
is the unit circle. Since the do-
main is the unit circle S
1
, we can Fourier-expand it! We can write
f(θ) =
X
nZ
ˆ
f
n
e
inθ
=
ˆ
f
0
+
X
n=1
ˆ
f
n
e
inθ
+
X
n=1
ˆ
f
n
e
inθ
.
However, we know that
z
=
re
and
¯z
=
re
. So on the boundary, we know
that z = e
and ¯z = e
. So we can write
f(θ) =
ˆ
f
0
+
X
n=1
(
ˆ
f
n
z
n
+
ˆ
f
n
¯z
n
)
This is defined for |z| = 1. Now we can define φ on the unit disk by
φ(z, ¯z) =
ˆ
f
0
+
X
n=1
ˆ
f
n
z
n
+
X
n=1
ˆ
f
n
¯z
n
.
It is clear that this obeys the boundary conditions by construction. Also,
φ
(
z, ¯z
)
certainly converges when
|z| <
1 if the series for
f
(
θ
) converged on
Ω. Also,
φ
(
z, ¯z
) is of the form
ψ
(
z
) +
χ
(
¯z
). So it is a solution to Laplace’s equation on
the unit disk. Since the solution is unique, we’re done!
4.3 Separation of variables
Unfortunately, the use of complex variables is very special to the case where
R
2
. In higher dimensions, we proceed differently.
We let =
{
(
x, y, z
)
R
3
: 0
x a,
0
y b, z
0
}
. We want the
following boundary conditions:
ψ(0, y, z) = ψ(a, y, z) = 0
ψ(x, 0, z) = ψ(x, b, z) = 0
lim
z→∞
ψ(x, y, z) = 0
ψ(x, y, 0) = f(x, y),
where
f
is a given function. In other words, we want our function to be
f
when
z = 0 and vanish at the boundaries.
The first step is to look for a solution of
2
ψ(x, y, z) = 0 of the form
ψ(x, y, z) = X(x)Y (y)Z(z).
Then we have
0 =
2
ψ = Y ZX
00
+ XZY
00
+ XY Z
00
= XY Z
X
00
X
+
Y
00
Y
+
Z
00
Z
.
As long as ψ 6= 0, we can divide through by ψ and obtain
X
00
X
+
Y
00
Y
+
Z
00
Z
= 0.
The key point is each term depends on only one of the variables (
x, y, z
). If we
vary, say,
x
, then
Y
00
Y
+
Z
00
Z
does not change. For the total sum to be 0,
X
00
X
must
be constant. So each term has to be separately constant. We can thus write
X
00
= λX, Y
00
= µY, Z
00
= (λ + µ)Z.
The signs before λ and µ are there just to make our life easier later on.
We can solve these to obtain
X = a sin
λx + b cos
λx,
Y = c sin
λy + d cos
λx,
Z = g exp(
p
λ + µz) + h exp(
p
λ + µz).
We now impose the homogeneous boundary conditions, i.e. the conditions that
ψ vanishes at the walls and at infinity.
At x = 0, we need ψ(0, y, z) = 0. So b = 0.
At x = a, we need ψ(a, y, z) = 0. So λ =
a
2
.
At y = 0, we need ψ(x, 0, z) = 0. So d = 0.
At y = b, we need ψ(x, b, z) = 0. So µ =
b
2
.
As z , ψ(x, y, z) 0. So g = 0.
Here we have n, m N.
So we found a solution
ψ(x, y, z) = A
n,m
sin
x
a
sin
y
b
exp(s
n,m
z),
where A
n,m
is an arbitrary constant, and
s
2
n,m
=
n
2
a
2
+
m
2
b
2
π
2
.
This obeys the homogeneous boundary conditions for any
n, m N
but not the
inhomogeneous condition at z = 0.
By linearity, the general solution obeying the homogeneous boundary condi-
tions is
ψ(x, y, z) =
X
n,m=1
A
n,m
sin
x
a
sin
y
b
exp(s
n,m
z)
The final step is to impose the inhomogeneous boundary condition
ψ
(
x, y,
0) =
f(x, y). In other words, we need
X
n,m=1
A
n,m
sin
x
a
sin
y
b
= f(x, y). ()
The objective is thus to find the A
n,m
coefficients.
We can use the orthogonality relation we’ve previously had:
Z
a
0
sin
kπx
a
sin
x
a
dx = δ
k,n
a
2
.
So we multiply () by sin
x
a
and integrate:
X
n,m=1
A
n,m
Z
a
0
sin
kπx
a
sin
x
a
dx sin
y
b
=
Z
a
0
sin
kπx
a
f(x, y) dx.
Using the orthogonality relation, we have
a
2
X
m=1
A
k,m
sin
y
b
=
Z
a
0
sin
kπx
a
f(x, y) dx.
We can perform the same trick again, and obtain
ab
4
A
k,j
=
Z
[0,a]×[0,b]
sin
kπx
a
sin
jπy
b
f(x, y) dx dy. ()
So we found the solution
ψ(x, y, z) =
X
n,m=1
A
n,m
sin
x
a
sin
y
b
exp(s
n,m
z)
where
A
m,n
is given by (
). Since we have shown that there is a unique solution
to Laplace’s equation obeying Dirichlet boundary conditions, we’re done.
Note that if we imposed a boundary condition at finite
z
, say 0
z c
,
then both sets of exponential
exp
(
±
λ + µz
) would have contributed. Similarly,
if
ψ
does not vanish at the other boundaries, then the
cos
terms would also
contribute.
In general, to actually find our
ψ
, we have to do the horrible integral for
A
m,n
, and this is not always easy (since integrals are hard ). We will just look at
a simple example.
Example. Suppose f(x, y) = 1. Then
A
m,n
=
4
ab
Z
a
0
sin
x
a
dx
Z
b
0
sin
y
b
dy =
(
16
π
2
mn
n, m both odd
0 otherwise
Hence we have
ψ(x, y, z) =
16
π
2
X
n,m odd
1
nm
sin
x
a
sin
y
b
exp(s
m,n
z).
4.4 Laplace’s equation in spherical polar coordinates
4.4.1 Laplace’s equation in spherical polar coordinates
We’ll often also be interested in taking
Ω = {(s, y, z) R
3
:
p
x
2
+ y
2
+ z
2
a}.
Since our domain is spherical, it makes sense to use some coordinate system
with spherical symmetry. We use spherical coordinates (r, θ, φ), where
x = r sin θ cos φ, y = r sin θ sin φ, z = r cos θ.
The Laplacian is
2
=
1
r
2
r
r
2
r
+
1
r
2
sin θ
θ
sin θ
θ
+
1
r
2
sin
2
θ
2
φ
2
.
Similarly, the volume element is
dV = dx dy dz = r
2
sin θ dr dθ dφ.
In this course, we’ll only look at axisymmetric solutions, where
ψ
(
r, θ, φ
) =
ψ
(
r, θ
)
does not depend on φ.
We perform separation of variables, where we look for any set of solution of
2
ψ = 0 inside where ψ(r, θ) = R(r)Θ(θ).
Laplace’s equation then becomes
ψ
r
2
1
R
d
dr
(r
2
R
0
) +
1
Θ sin θ
d
dθ
sin θ
dθ

= 0.
Similarly, since each term depends only on one variable, each must separately
be constant. So we have
d
dr
r
2
dR
dr
= λR,
d
dθ
sin θ
dθ
= λ sin θΘ.
Note that both these equations are eigenfunction equations for some Sturm-
Liouville operator and the Θ(
θ
) equation has a non-trivial weight function
w(θ) = sin θ.
4.4.2 Legendre Polynomials
The angular equation is known as the Legendre’s equation. It’s standard to set
x
=
cos θ
(which is not the Cartesian coordinate, but just a new variable). Then
we have
d
dθ
= sin θ
d
dx
.
So the Legendre’s equation becomes
sin θ
d
dx
sin θ(sin θ)
dx
+ λ sin θΘ = 0.
Equivalently, in terms of x, we have
d
dx
(1 x
2
)
dx
= λΘ.
Note that since 0 θ π, the domain of x is 1 x 1.
This operator is a Sturm–Liouville operator with
p(x) = 1 x
2
, q(x) = 0.
For the Sturm–Liouville operator to be self-adjoint, we had
(g, Lf ) = (Lg, f) + [p(x)(g
f
0
g
0∗
f)]
1
1
.
We want the boundary term to vanish. Since
p
(
x
) = 1
x
2
vanishes at our
boundary
x
=
±
1, the Sturm-Liouville operator is self-adjoint provided our
function Θ(
x
) is regular at
x
=
±
1. Hence we look for a set of Legendre’s
equations inside (
1
,
1) that remains regular including at
x
=
±
1. We can try a
power series
Θ(x) =
X
n=0
a
n
x
n
.
The Legendre’s equation becomes
(1 x
2
)
X
n=0
a
n
n(n 1)x
n2
2
X
n=0
a
n
nx
n
+ λ
X
n=0
a
n
x
n
= 0.
For this to hold for all
x
(
1
,
1), the equation must hold for each coefficient
of x separately. So
a
n
(λ n(n + 1)) + a
n2
(n + 2)(n + 1) = 0.
This requires that
a
n+2
=
n(n + 1) λ
(n + 2)(n + 1)
a
n
.
This equation relates
a
n+2
to
a
n
. So we can choose
a
0
and
a
1
freely. So we get
two linearly independents solutions Θ
0
(x) and Θ
1
(x), where they satisfy
Θ
0
(x) = Θ
0
(x), Θ
1
(x) = Θ
1
(x).
In particular, we can expand our recurrence formula to obtain
Θ
0
(x) = a
0
1
λ
2
x
2
(6 λ)λ
4!
x
4
+ ···
Θ
1
(x) = a
1
x +
(2 λ)
3!
x
2
+
(12 λ)(2 λ)
5!
x
5
+ ···
.
We now consider the boundary conditions. We know that Θ(
x
) must be regular
(i.e. finite) at x = ±1. This seems innocent, However, we have
lim
n→∞
a
n+2
a
n
= 1.
This is fine if we are inside (
1
,
1), since the power series will still converge.
However, at
±
1, the ratio test is not conclusive and does not guarantee conver-
gence. In fact, more sophisticated convergence tests show that the infinite series
would diverge on the boundary! To avoid this problem, we need to choose
λ
such
that the series truncates. That is, if we set
λ
=
`
(
`
+ 1) for some
` N
, then
our power series will truncate. Then since we are left with a finite polynomial,
it of course converges.
In this case, the finiteness boundary condition fixes our possible values of
eigenvalues. This is how quantization occurs in quantum mechanics. In this case,
this process is why angular momentum is quantized.
The resulting polynomial solutions
P
`
(
x
) are called Legendre polynomials.
For example, we have
P
0
(x) = 1
P
1
(x) = x
P
2
(x) =
1
2
(3x
2
1)
P
3
(x) =
1
2
(5x
3
3x),
where the overall normalization is chosen to fix P
`
(1) = 1.
It turns out that
P
`
(x) =
1
2
`
`!
d
`
dx
`
(x
2
1)
`
.
The constants in front are just to ensure normalization. We now check that this
indeed gives the desired normalization:
P
`
(1) =
1
2
`
d
`
dx
`
[(x 1)
`
(x + 1)
`
]
x=1
=
1
2
`
`!
`!(x + 1)
`
+ (x 1)(stuff)
x=1
= 1
We have previously shown that the eigenfunctions of Sturm-Liouville operators
are orthogonal. Let’s see that directly now for this operator. The weight function
is w(x) = 1.
We first prove a short lemma: for 0 k `, we have
d
k
dx
k
(x
2
1)
`
= Q
`,k
(x
2
1)
`k
for some degree k polynomial Q
`,k
(x).
We show this by induction. This is trivially true when k = 0. We have
d
k+1
dx
k+1
(x
2
+ 1)
`
=
d
dx
Q
`,k
(x)(x
2
1)
`k
= Q
0
`,k
(x
2
1)
`k
+ 2(` k)Q
`,k
x(x
2
1)
`k1
= (x
2
1)
`k1
h
Q
0
`,k
(x
2
1) 2(` k)Q
`,k
x
i
.
Since
Q
`,k
has degree
k
,
Q
0
`,k
has degree
k
1. So the right bunch of stuff has
degree k + 1.
Now we can show orthogonality. We have
(P
`
, P
`
0
) =
Z
1
1
P
`
(x)P
`
0
(x) dx
=
1
2
`
`!
Z
1
1
d
`
dx
`
(x
2
1)
`
P
`
0
(x) dx
=
1
2
`
`!
d
`1
dx
`1
(x
2
1)
2
P
`
0
(x)
1
2
`
`!
Z
1
1
d
`1
dx
`1
(x
2
1)
`
dP
`
0
dx
dx
=
1
2
`
`!
Z
1
1
d
`1
dx
`1
(x
2
1)
`
dP
`
0
dx
dx.
Note that the first term disappears since the (
`
1)
th
derivative of (
x
2
1)
`
still
has a factor of
x
2
1. So integration by parts allows us to transfer the derivative
from x
2
1 to P
`
0
.
Now if
` 6
=
`
0
, we can wlog assume
`
0
< `
. Then we can integrate by parts
`
0
times until we get the `
0th
derivative of P
`
0
, which is zero.
In fact, we can show that
(P
`
, P
`
0
) =
2δ
``
0
2` + 1
.
hence the P
`
(x) are orthogonal polynomials.
By the fundamental theorem of algebra, we know that
P
`
(
x
) has
`
roots. In
fact, they are always real, and lie in (1, 1).
Suppose not. Suppose only
m < `
roots lie in (
1
,
1). Then let
Q
m
(
x
) =
Q
m
r=1
(
xx
r
), where
{x
1
, x
2
, ··· , x
m
}
are these
m
roots. Consider the polynomial
P
`
(
x
)
Q
m
(
x
). If we factorize this, we get
Q
`
r=m+1
(
xr
)
Q
m
r=1
(
xx
r
)
2
. The first
few terms have roots outside (
1
,
1), and hence do not change sign in (
1
,
1).
The latter terms are always non-negative. Hence for some appropriate sign, we
have
±
Z
1
1
P
`
(x)Q
m
(x) dx > 0.
However, we can expand
Q
m
(x) =
m
X
r=1
q
r
P
r
(x),
but (P
`
, P
r
) = 0 for all r < `. This is a contradiction.
4.4.3 Solution to radial part
In our original equation
2
φ
= 0, we now set Θ(
θ
) =
P
`
(
cos θ
). The radial
equation becomes
(r
2
R
0
)
0
= `(` + 1)R.
Trying a solution of the form
R
(
r
) =
r
α
, we find that
α
(
α
+ 1) =
`
(
`
+ 1). So
the solution is α = ` or (` + 1). Hence
φ(r, θ) =
X
a
`
r
`
+
b
`
r
`+1
P
`
(cos θ)
is the general solution to the Laplace equation
2
φ
= 0 on the domain. In we
want regularity at r = 0, then we need b
`
= 0 for all `.
The a
`
are fixed if we require φ(a, θ) = f(θ) on Ω, i.e. when r = 1.
We expand
f(θ) =
X
`=0
F
`
P
`
(cos θ),
where
F
`
= (P
`
, f) =
Z
π
0
P
`
(cos θ)f(θ) d(cos θ).
So
φ(r, θ) =
X
`=0
F
`
r
a
α
P
`
(cos θ).
4.5 Multipole expansions for Laplace’s eq uation
One can quickly check that
φ(r) =
1
|r r
0
|
solves Laplace’s equation
2
φ
= 0 for all
r R
3
\ r
0
. For example, if
r
0
=
ˆ
k
,
where
ˆ
k is the unit vector in the z direction, then
1
|r
ˆ
k|
=
1
r
2
+ 1 2r cos θ
=
X
`=0
c
`
r
`
P
`
(cos θ).
To find these coefficients, we can employ a little trick. Since
P
`
(1) = 0, at
θ
= 0,
we have
X
`=0
c
`
r
`
=
1
1 r
=
X
`=0
r
`
.
So all the coefficients must be 1. This is valid for r < 1.
More generally, we have
1
|r r
0
|
=
1
r
0
X
`=0
r
r
0
`
P
`
(
ˆ
r ·
ˆ
r
0
).
This is called the multiple expansion, and is valid when r < r
0
. Thus
1
|r r
0
|
=
1
r
0
+
r
r
02
ˆ
r ·
ˆ
r
0
+ ··· .
The first term
1
r
0
is known as the monopole, and the second term is known as
the dipole, in analogy to charges in electromagnetism. The monopole term is
what we get due to a single charge, and the second term is the result of the
interaction of two charges.
4.6 Laplace’s equation in cylindrical coordinates
We let
Ω = {(r, θ, z) R
3
: r a, z 0}.
In cylindrical polars, we have
2
φ =
1
r
r
r
φ
r
+
1
r
2
2
φ
θ
2
+
2
φ
z
2
= 0.
We look for a solution that is regular inside Ω and obeys the boundary conditions
φ(a, θ, z) = 0
φ(r, θ, 0) = f(r, θ)
lim
z→∞
φ(r, θ, z) = 0,
where f is some fixed function.
Again, we start by separation of variables. We try
φ(r, θ, z) = R(r)Θ(θ)Z(z).
Then we get
1
rR
(rR
0
)
0
+
1
r
2
Θ
00
Θ
+
Z
00
Z
= 0.
So we immediately know that
Z
00
= µZ
We now replace Z
00
/Z by µ, and multiply by r
2
to obtain
r
R
(rR
0
)
0
+
Θ
00
Θ
+ µr
2
= 0.
So we know that
Θ
00
= λΘ.
Then we are left with
r
2
R
00
+ rR
0
+ (µr
2
λ)R = 0.
Since we require periodicity in θ, we have
Θ(θ) = a
n
sin + b
n
cos , λ = n
2
, n N.
Since we want the solution to decay as z , we have
Z(z) = c
µ
e
µz
The radial equation is of Sturm-Liouville type since it can be written as
d
dr
r
dR
dr
n
2
r
R = µrR.
Here we have
p(r) = r, q(r) =
n
2
r
, w(r) = r.
Introducing x = r
µ, we can rewrite this as
x
2
d
2
R
dx
2
+ x
dR
dx
+ (x
2
n
2
)R = 0.
This is Bessel’s equation. Note that this is actually a whole family of differential
equations, one for each
n
. The
n
is not the eigenvalue we are trying to figure out
in this equation. It is already fixed by the Θ equation. The actual eigenvalue we
are finding out is µ, not n.
Since Bessel’s equations are second-order, there are two independent solution
for each n, namely J
n
(x) and Y
n
(x). These are called Bessel functions of order
n
of the first (
J
n
) or second (
Y
n
) kind. We will not study these functions’
properties, but just state some useful properties of them.
The
J
n
’s are all regular at the origin. In particular, as
x
0
J
n
(
x
)
x
n
.
These look like decaying sine waves, but the zeroes are not regularly spaced.
On the other hand, the
Y
n
’s are similar but singular at the origin. As
x
0,
Y
0
(x) ln x, while Y
n
(x) x
n
.
Now we can write our separable solution as
φ(r, θ, z) = (a
n
sin + b
n
cos )e
µz
[c
µ,n
J
n
(r
µ) + d
µ,n
Y
n
(r,
µ)].
Since we want regularity at
r
= 0, we don’t want to have the
Y
n
terms. So
d
µ,n
= 0.
We now impose our first boundary condition
φ
(
a, θ, z
) = 0. This demands
J
n
(a
µ) = 0. So we must have
µ =
k
ni
a
,
where
k
ni
is the
i
th root of
J
n
(
x
) (since there is not much pattern to these roots,
this is the best description we can give!).
So our general solution obeying the homogeneous conditions is
φ(r, θ, z) =
X
n=0
X
iroots
(A
ni
sin + B
ni
cos ) exp
k
ni
a
z
J
n
k
ni
r
a
.
We finally impose the inhomogeneous boundary condition
φ
(
r, θ,
0) =
f
(
r, θ
). To
do this, we use the orthogonality condition
Z
a
0
J
n
k
nj
r
a
J
n
k
ni
r
a
r dr =
a
2
2
δ
ij
[J
0
n
(k
ni
)]
2
,
which is proved in the example sheets. Note that we have a weight function
r
here. Also, remember this is the orthogonality relation for different roots of
Bessel’s functions of the same order. It does not relate Bessel’s functions of
different orders.
We now integrate our general solution with respect to cos to obtain
1
π
Z
π
π
f(r, θ) cos dθ =
X
iroots
J
m
k
mi
r
a
.
Now we just have Bessel’s function for a single order
j
. So we can use the
orthogonality relation for the Bessel’s functions to obtain
B
mj
=
1
[J
0
m
(k
mj
)]
2
πa
2
Z
a
0
Z
π
π
cos J
m
k
mj
r
a
f(r, θ)r dr dθ.
How can we possibly perform this integral? We don’t know how
J
m
looks like,
what the roots
k
mj
are, and we multiply all these complicated things together
and integrate! Often, we just don’t. We ask our computers to do this numerically
for us. There are indeed some rare cases where we are indeed able to get an
explicit solution, but these are rare.
4.7 The heat equation
We choose =
R
n
×
[0
,
). We treat
R
n
as the “space” variable, and [0
,
) as
our time variable.
Definition (Heat equation). The heat equation for a function φ : Ω C is
φ
t
= κ
2
φ,
where κ > 0 is the diffusion constant.
We can think of this as a function
φ
representing the “temperature” at
different points in space and time, and this equation says the rate of change of
temperature is proportional to
2
φ.
Our previous study of the Laplace’s equation comes in handy. When our
system is in equilibrium and does not change with time, then we are left with
Laplace’s equation.
Let’s first look at some interesting properties of the heat equation.
The first most important property of the heat equation is that the total
“heat” is conserved under evolution by the heat equation, i.e.
d
dt
Z
R
n
φ(x, t) d
n
x = 0,
since we can write this as
d
dt
Z
R
n
φ(x, t) d
n
x =
Z
R
n
φ
t
d
n
x = κ
Z
R
n
2
φ d
n
x = 0,
provided φ(x, t) 0 as |x| .
In particular, on a compact space, i.e. =
M ×
[0
,
) with
M
compact,
there is no boundary term and we have
d
dt
Z
M
φ dµ = 0.
The second useful property is if
φ
(
x, t
) solves the heat equation, so do
φ
1
(
x, t
) =
φ(x x
0
, t t
0
), and φ
2
(x, t) = (λx, λ
2
t).
Let’s try to choose A such that
Z
R
n
φ
2
(x, t) d
n
x =
Z
R
n
φ(x, t) d
n
x.
We have
Z
R
n
φ
2
(x, t) d
n
x = A
Z
R
n
φ(λx, λ
2
t) d
n
x =
n
Z
R
n
φ(y, λ
2
t) d
n
y,
where we substituted y = λx.
So if we set
A
=
λ
n
, then the total heat in
φ
2
at time
λ
2
t
is the same as the
total heat in
φ
at time
t
. But the total heat is constant in time. So the total
heat is the same all the time.
Note again that if
φ
t
=
κ
2
φ
on Ω
×
[0
,
), then so does
λ
n
φ
(
λx, λ
2
t
). This
means that nothing is lost by letting
λ
=
1
and consider solutions of the form
φ(x, t) =
1
(κt)
n/2
F
x
κt
=
1
(κt)
n/2
F (η),
where
η =
x
κt
.
For example, in 1 + 1 dimensions, we can look for a solution of the form
1
κt
F
(
η
).
We have
φ
t
=
t
1
κt
F (η)
=
1
2
κt
3
F (η) +
1
κt
dη
dt
F
0
(η) =
1
2
κt
3
[F + ηF
0
]
On the other side of the equation, we have
κ
2
x
2
1
κt
F (η)
=
κ
κt
x
η
x
F
0
=
1
κt
3
F
00
.
So the equation becomes
0 = 2F
00
+ ηF
0
+ F = (2F
0
+ ηF )
0
.
So we have
2F
0
+ ηF = const,
and the boundary conditions require that the constant is zero. So we get
F
0
=
η
2
F.
We now see a solution
F (η) = a exp
η
2
4
.
By convention, we now normalize our solution φ(x, t) by requiring
Z
−∞
φ(x, t) dx = 1.
This gives
a =
1
4π
,
and the full solution is
φ(x, t) =
1
4πκt
exp
x
2
4κt
.
More generally, on R
n
× [0, ), we have the solution
φ(x, t) =
1
(4π(t t
0
))
n/2
exp
(x x
0
)
2
4κ(t t
0
)
.
At any fixed time
t 6
=
t
0
, the solutions are Gaussians centered on
x
0
with standard
deviation
p
2κ(t t
0
), and the height of the peak at x = x
0
is
q
1
4πκ(tt
0
)
.
We see that the solution gets “flatter” as
t
increases. This is a general property
of the evolution by the heat equation. Suppose we have an eigenfunction
y
(
x
) of
the Laplacian on Ω, i.e.
2
y
=
λy
. We want to show that in certain cases,
λ
must be positive. Consider
λ
Z
|y|
2
d
n
x =
Z
y
(x)
2
y(x) d
n
x =
Z
y
n · y d
n1
x
Z
|∇y|
2
d
n
x.
If there is no boundary term, e.g. when is closed and compact (e.g. a sphere),
then we get
λ =
R
|∇y|
2
d
n
x
R
|y|
2
d
n
x
.
Hence the eigenvalues are all positive. What does this mean for heat flow?
Suppose
φ
(
x, t
) obeys the heat equation for
t >
0. At time
t
= 0, we have
φ(x, 0) = f(x). We can write the solution (somehow) as
φ(x, t) = e
t
2
φ(x, 0),
where we (for convenience) let κ = 1. But we can expand our initial condition
φ(x, 0) = f(x) =
X
I
c
I
y
I
(x)
in a complete set {y
I
(x)} of eigenfunctions for
2
.
By linearity, we have
φ(x, t) = e
t
2
X
I
c
I
y
I
(x)
!
=
X
I
c
I
e
t
2
y
I
(x)
=
X
I
c
I
e
λ
2
t
y
I
(x)
= c
I
(t)y
I
(x),
where
c
I
(
t
) =
c
I
e
λ
2
t
. Since
λ
I
are all positive, we know that
c
I
(
t
) decays
exponentially with time. In particular, the coefficients corresponding to the
largest eigenvalues
|λ
I
|
decays most rapidly. We also know that eigenfunctions
with larger eigenvalues are in general more “spiky”. So these smooth out rather
quickly.
When people first came up with the heat equation to describe heat loss, they
were slightly skeptical this flattening out caused by the heat equation is not
time-reversible, while Newton’s laws of motions are all time reversible. How
could this smoothening out arise from reversible equations?
Einstein came up with an example where the heat equation can come out of
apparently naturally from seemingly reversible laws, namely Brownian motion.
The idea is that a particle will randomly jump around in space, where the
movement is independent of time and position.
Let the probability that a dust particle moves through a step
y
over a time
t be p(y, t). For any fixed t, we must have
Z
−∞
p(y, t) dy = 1.
We also assume that
p
(
y,
t
) is independent of time, and of the location of the
dust particle. We also assume
p
(
y,
t
) is strongly peaked around
y
= 0 and
p
(
y,
t
) =
p
(
y,
t
). Now let
P
(
x, t
) be the probability that the dust particle
is located at x at time t. Then at time t + δt, we have
P (x, t + t) =
Z
−∞
P (x y, t)p(y, t) dy.
For P (x y, t) sufficiently smooth, we can write
P (x y, t) P (x, t) y
P
x
(x, t) +
y
2
2
2
P
x
2
(x, t).
So we get
P (x, t + t) P (x, t)
P
x
(x, t)hyi+
1
2
hy
2
i
2
P
x
2
P (x, t) + ··· ,
where
hy
r
i =
Z
−∞
y
r
p(y, t) dy.
Since
p
is even, we expect
hy
r
i
to vanish when
r
is odd. Also, since
y
is strongly
peaked at 0, we expect the higher order terms to be small. So we can write
P (x, t + t) P (x, t) =
1
2
hy
2
i
2
P
x
2
.
Suppose that as we take the limit
t
0, we get
hy
2
i
2∆t
κ
for some
κ
. Then
this becomes the heat equation
P
t
= κ
2
P
x
2
.
Proposition.
Suppose
φ
: Ω
×
[0
,
)
R
satisfies the heat equation
φ
t
=
κ
2
φ
,
and obeys
Initial conditions φ(x, 0) = f(x) for all x
Boundary condition φ(x, t)|
= g(x, t) for all t [0, ).
Then φ(x, t) is unique.
Proof. Suppose φ
1
and φ
2
are both solutions. Then define Φ = φ
1
φ
2
and
E(t) =
1
2
Z
Φ
2
dV.
Then we know that
E
(
t
)
0. Since
φ
1
, φ
2
both obey the heat equation, so does
Φ. Also, on the boundary and at t = 0, we know that Φ = 0. We have
dE
dt
=
Z
Φ
dt
dV
= κ
Z
Φ
2
Φ dV
= κ
Z
ΦΦ · dS κ
Z
(Φ)
2
dV
= κ
Z
(Φ)
2
dV
0.
So we know that
E
decreases with time but is always non-negative. We also
know that at time
t
= 0,
E
= Φ = 0. So
E
= 0 always. So Φ = 0 always. So
φ
1
= φ
2
.
Example
(Heat conduction in uniform medium)
.
Suppose we are on Earth,
and the Sun heats up the surface of the Earth through sun light. So the sun will
maintain the soil at some fixed temperature. However, this temperature varies
with time as we move through the day-night cycle and seasons of the year.
We let
φ
(
x, t
) be the temperature of the soil as a function of the depth
x
,
defined on R
0
× [0, ). Then it obeys the heat equation with conditions
(i) φ(0, t) = φ
0
+ A cos
2πt
t
D
+ B cos
2πt
t
Y
.
(ii) φ(x, t) const as x .
We know that
φ
t
= K
2
φ
x
2
.
We try the separation of variables
φ(x, t) = T (t)X(x).
Then we get the equations
T
0
= λT, X
00
=
λ
K
X.
From the boundary solution, we know that our things will be oscillatory. So we
let λ be imaginary, and set λ = . So we have
φ(x, t) = e
t
a
ω
e
/Kx
+ b
ω
e
/Kx
.
Note that we have
r
K
=
(1 + i)
q
|ω|
2K
ω > 0
(i 1)
q
|ω|
2K
ω < 0
Since
φ
(
x, t
)
constant as
x
, we don’t want our
φ
to blow up. So if
ω <
0,
we need a
ω
= 0. Otherwise, we need b
ω
= 0.
To match up at x = 0, we just want terms with
|ω| = ω
D
=
2π
t
D
, |ω| = ω
Y
=
2π
t
Y
.
So we can write out solution as
φ(x, t) = φ
0
+ A exp
r
ω
D
2K
cos
ω
D
t
r
ω
D
2K
x
+ B exp
r
ω
Y
2K
cos
ω
Y
t
r
ω
Y
2K
x
We can notice a few things. Firstly, as we go further down, the effect of the sun
decays, and the temperature is more stable. Also, the effect of the day-night
cycle decays more quickly than the annual cycle, which makes sense. We also
see that while the temperature does fluctuate with the same frequency as the
day-night/annual cycle, as we go down, there is a phase shift. This is helpful
since we can store things underground and make them cool in summer, warm in
winter.
Example
(Heat conduction in a finite rod)
.
We now have a finite rod of length
2L.
x = 0
x = L x = L
We have the initial conditions
φ(x, 0) = Θ(x) =
(
1 0 < x < L
0 L < x < 0
and the boundary conditions
φ(L, t) = 0, φ(L, t) = 1
So we start with a step temperature, and then maintain the two ends at fixed
temperatures 0 and 1.
We are going to do separation of variables, but we note that all our boundary
and initial conditions are inhomogeneous. This is not helpful. So we use a
little trick. We first look for any solution satisfying the boundary conditions
φ
S
(
L, t
) = 0,
φ
S
(
L, t
) = 1. For example, we can look for time-independent
solutions φ
S
(x, t) = φ
S
(x). Then we need
d
2
φ
S
dx
2
= 0. So we get
φ
S
(x) =
x + L
2L
.
By linearity,
ψ
(
x, t
) =
φ
(
x, t
)
φ
s
(
x
) obeys the heat equation with the conditions
ψ(L, t) = ψ(L, t) = 0,
which is homogeneous! Our initial condition now becomes
ψ(x, 0) = Θ(x)
x + L
2L
.
We now perform separation of variables for
ψ(x, t) = X(x)T (t).
Then we obtain the equations
T
0
= κλT, X
0
= λX.
Then we have
ψ(x, t) =
h
a sin(
λx) + b cos(
λx)
i
e
κλt
.
Since initial condition is odd, we can eliminate all
cos
terms. Our boundary
conditions also requires
λ =
n
2
π
2
L
2
, n = 1, 2, ··· .
So we have
φ(x, t) = φ
s
(x) +
X
n=1
a
n
sin
x
L
exp
κn
2
π
2
L
2
t
,
where a
n
are the Fourier coefficients
a
n
=
1
L
Z
L
L
Θ(x)
x + L
2L
sin
x
L
dx =
1
.
Example
(Cooling of a uniform sphere)
.
Once upon a time, Charles Darwin
went around the Earth, looked at species, and decided that evolution happened.
When he came up with his theory, it worked quite well, except that there was
one worry. Was there enough time on Earth for evolution to occur?
This calculation was done by Kelvin. He knew well that the Earth started as
a ball of molten rock, and obviously life couldn’t have evolved when the world
was still molten rock. So he would want to know how long it took for the Earth
to cool down to its current temperature, and if that was sufficient for evolution
to occur.
We can, unsurprisingly, use the heat equation. We assume that at time
t
= 0,
the temperature is
φ
0
, the melting temperature of rock. We also assume that
the space is cold, and we let the temperature on the boundary of Earth as 0. We
then solve the heat equation on a sphere (or ball), and see how much time it
takes for Earth to cool down to its present temperature.
We let Ω =
{
(
x, y, z
)
R
3
, r R}
, and we want a solution
φ
(
r, t
) of the heat
equation that is spherically symmetric and obeys the conditions
φ(R, t) = 0
φ(r, 0) = φ
0
,
We can write the heat equation as
φ
t
= κ
2
φ =
κ
r
2
r
r
2
φ
r
.
Again, we do the separation of variables.
φ(r, t) = R(r)T (t).
So we get
T
0
= λ
2
κT,
d
dr
r
2
dR
dr
= λ
2
r
2
R.
We can simplify this a bit by letting
R
(
r
) =
S(r)
r
, then our radial equation just
becomes
S
00
= λ
2
S.
We can solve this to get
R(r) = A
λ
sin λr
r
+ B
λ
cos λr
r
.
We want a regular solution at
r
= 0. So we need
B
λ
= 0. Also, the boundary
condition φ(R, t) = 0 gives
λ =
R
, n = 1, 2, ···
So we get
φ(r, t) =
X
nZ
A
n
r
sin
r
R
exp
κn
2
π
2
t
R
2
,
where our coefficients are
A
n
= (1)
n+1
φ
0
R
.
We know that the Earth isn’t just a cold piece of rock. There are still volcanoes.
So we know many terms still contribute to the sum nowadays. This is rather
difficult to work with. So we instead look at the temperature gradient
φ
r
=
φ
0
r
X
nZ
(1)
n+1
cos
r
R
exp
κn
2
π
2
t
R
2
+ sin stuff.
We evaluate this at the surface of the Earth, R = r. So we get the gradient
φ
r
R
=
φ
0
R
X
nZ
exp
κn
2
π
2
t
R
2
φ
0
R
Z
−∞
exp
κπy
2
t
R
2
dy = φ
0
r
1
πκt
.
So the age of the Earth is approximately
t
φ
0
V
2
1
4πK
,
where
V =
φ
r
R
.
We can plug the numbers in, and get that the Earth is 100 million years. This is
not enough time for evolution.
Later on, people discovered that fission of metals inside the core of Earth
produce heat, and provides an alternative source of heat. So problem solved!
The current estimate of the age of the universe is around 4 billion years, and
evolution did have enough time to occur.
4.8 The wave equation
Consider a string x [0, L] undergoing small oscillations described by φ(x, t).
φ(x, t)
A
B
θ
A
θ
B
Consider two points
A
,
B
separated by a small distance
δx
. Let
T
A
(
T
B
) be
the outward pointing tension tangent to the string at
A
(
B
). Since there is no
sideways (x) motion, there is no net horizontal force. So
T
A
cos θ
A
= T
B
cos θ
B
= T. ()
If the string has mass per unit length
µ
, then in the vertical direction, Newton’s
second law gives
µδx
2
φ
t
2
= T
B
sin θ
B
T
A
sin θ
A
.
We now divide everything by T , noting the relation (), and get
µ
δx
T
2
φ
t
2
=
T
B
sin θ
B
T
B
cos θ
B
T
A
sin θ
A
T
A
cos θ
A
= tan θ
B
tan θ
A
=
φ
x
B
φ
x
A
δx
2
φ
x
2
.
Taking the limit
δx
0 and setting
c
2
=
T
, we have that
φ
(
x, t
) obeys the
wave equation
1
c
2
2
φ
t
2
=
2
φ
x
2
.
From IA Differential Equations, we’ve learnt that the solution can be written as
f
(
x ct
) +
g
(
x
+
ct
). However, this does not extend to higher dimensions, but
separation of variables does. So let’s do that.
Assume that the string is fixed at both ends. Then
φ
(0
, t
) =
φ
(
L, t
) = 0 for
all
t
. The we can perform separation of variables, and the general solution can
then be written
φ(x, t) =
X
n=1
sin
x
L
A
n
cos
ct
L
+ B
n
sin
ct
L

.
The coefficients
A
n
are fixed by the initial profile
φ
(
x,
0) of the string, while
the coefficients
B
n
are fixed by the initial string velocity
φ
t
(
x,
0). Note that we
need two sets of initial conditions, since the wave equation is second-order in
time.
Energetics and uniqueness
An oscillating string contains has some sort of energy. The kinetic energy of a
small element δx of the string is
1
2
µδx
φ
t
2
.
The total kinetic energy of the string is hence the integral
K(t) =
µ
2
Z
L
0
φ
t
2
dx.
The string also has potential energy due to tension. The potential energy of a
small element δx of the string is
T × extension = T (δs δx)
= T (
p
δx
2
+ δφ
2
δx)
T δx
1 +
1
2
δφ
δx
2
+ ···
!
T δx
=
T
2
δx
δφ
δx
2
.
Hence the total potential energy of the string is
V (t) =
µ
2
Z
L
0
c
2
φ
x
2
dx,
using the definition of c
2
.
Using our series expansion for φ, we get
K(t) =
µπ
2
c
2
4L
X
n=1
n
2
A
n
sin
ct
L
B
n
cos
ct
L

2
V (t) =
µπ
2
c
2
4L
X
n=1
n
2
A
n
cos
ct
L
+ B
n
sin
ct
L

2
The total energy is
E(t) =
2
c
2
4L
X
n=1
n
2
(A
2
n
+ B
2
n
).
What can we see here? Our solution is essentially an (infinite) sum of independent
harmonic oscillators, one for each
n
. The period of the fundamental mode (
n
= 1)
is
2π
ω
= 2
π ·
L
πc
=
2L
c
. Thus, averaging over a period, the average kinetic energy
is
¯
K =
c
2L
Z
2L/c
0
K(t) dt =
¯
V =
c
2L
Z
2L/c
0
V (t) dt =
E
2
.
Hence we have an equipartition of the energy between the kinetic and potential
energy.
The energy also allows us to prove a uniqueness theorem. First we show that
energy is conserved in general.
Proposition.
Suppose
φ
: Ω
×
[0
,
)
R
obeys the wave equation
2
φ
t
2
=
c
2
2
φ
inside × (0, ), and is fixed at the boundary. Then E is constant.
Proof. By definition, we have
dE
dt
=
Z
2
ψ
t
2
ψ
t
+ c
2
φ
t
· φ dV.
We integrate by parts in the second term to obtain
dE
dt
=
Z
dφ
dt
2
φ
t
2
c
2
2
φ
dV + c
2
Z
φ
t
φ · dS.
Since
2
φ
t
2
=
c
2
2
φ
by the wave equation, and
φ
is constant on
Ω, we know
that
dE
φ
dt
= 0.
Proposition.
Suppose
φ
: Ω
×
[0
,
)
R
obeys the wave equation
2
φ
t
2
=
c
2
2
φ
inside × (0, ), and obeys, for some f, g, h,
(i) φ(x, 0) = f (x);
(ii)
φ
t
(x, 0) = g(x); and
(iii) φ|
×[0,)
= h(x).
Then φ is unique.
Proof.
Suppose
φ
1
and
φ
2
are two such solutions. Then
ψ
=
φ
1
φ
2
obeys the
wave equation
2
ψ
t
2
= c
2
2
ψ,
and
ψ|
×[0,)
= ψ|
×{0}
=
ψ
t
×{0}
= 0.
We let
E
ψ
(t) =
1
2
Z
"
ψ
t
2
+ c
2
ψ · ψ
#
dV.
Then since
ψ
obeys the wave equation with fixed boundary conditions, we know
E
ψ
is constant.
Initially, at
t
= 0, we know that
ψ
=
ψ
t
= 0. So
E
ψ
(0) = 0. At time
t
, we
have
E
ψ
=
1
2
Z
ψ
t
2
+ c
2
(ψ) · (ψ) dV = 0.
Hence we must have
ψ
t
= 0. So
ψ
is constant. Since it is 0 at the beginning, it
is always 0.
Example. Consider Ω = {(x, y) R
2
, x
2
+ y
2
1}, and let φ(r, θ, t) solve
1
c
2
2
φ
t
2
=
2
φ =
1
r
r
φ
r
+
1
r
2
2
φ
θ
2
,
with the boundary condition
φ|
= 0. We can imagine this as a drum, where
the membrane can freely oscillate with the boundary fixed.
Separating variables with φ(r, θ, t) = T (t)R(r)Θ(θ), we get
T
00
= c
2
λT, Θ
00
= µΘ, r(R
0
)
0
+ (r
2
λ µ)R = 0.
Then as before,
T
and Θ are both sine and cosine waves. Since we are in polars
coordinates, we need
φ
(
t, r, θ
+ 2
π
) =
φ
(
t, r, θ
). So we must have
µ
=
m
2
for
some m N. Then the radial equation becomes
r(rR
0
)
0
+ (r
2
λ m
2
)R = 0,
which is Bessel’s equation of order m. So we have
R(r) = a
m
J
m
(
λr) + b
m
Y
m
(
λr).
Since we want regularity at
r
= 0, we need
b
m
= 0 for all
m
. To satisfy the
boundary condition
φ|
= 0, we must choose
λ
=
k
mi
, where
k
mi
is the
i
th
root of J
m
.
Hence the general solution is
φ(t, r, θ) =
X
i=0
[A
0i
sin(k
0i
ct) + B
0i
cos(k
0i
ct)]J
0
(k
0i
r)
+
X
m=1
X
i=0
[A
mi
cos() + B
mi
sin()] sin k
mi
ctJ
m
(k
mi
r)
+
X
m=1
X
i=0
[C
mi
cos() + D
mi
sin()] cos k
mi
ctJ
m
(k
mi
r)
For example, suppose we have the initial conditions
φ
(0
, r, θ
) = 0,
t
φ
(0
, r, θ
) =
g
(
r
). So we start with a flat surface and suddenly hit it with some force. By
symmetry, we must have
A
mi
, B
mi
, C
mi
, D
mi
= 0 for
m 6
= 0. If this does not
seem obvious, we can perform some integrals with the orthogonality relations to
prove this.
At t = 0, we need φ = 0. So we must have B
0i
= 0. So we are left with
φ =
X
i=0
A
0i
sin(k
0i
ct)J
0
(k
0j
r).
We can differentiate this, multiply with J
0
(k
0j
r)r to obtain
Z
1
0
X
i=0
k
0i
cA
0i
J
0
(k
0i
r)J
0
(k
0j
r)r dr =
Z
1
0
g(r)J
0
(k
0j
r)r dr.
Using the orthogonality relations for J
0
from the example sheet, we get
A
0i
=
2
ck
0i
1
[J
0
0
(k
0i
)]
2
Z
1
0
g(r)J
0
(k
0j
r)r dr.
Note that the frequencies come from the roots of the Bessel’s function, and are
not evenly spaced. This is different from, say, string instruments, where the
frequencies are evenly spaced. So drums sound differently from strings.
So far, we have used separation of variables to solve our differential equations.
This is quite good, as it worked in our examples, but there are a few issues with
it. Of course, we have the problem of whether it converges or not, but there is a
more fundamental problem.
To perform separation of variables, we need to pick a good coordinate system,
such that the boundary conditions come in a nice form. We were able to
perform separation of variables because we can find some coordinates that fit our
boundary conditions well. However, in real life, most things are not nice. Our
domain might have a weird shape, and we cannot easily find good coordinates
for it.
Hence, instead of solving things directly, we want to ask some more general
questions about them. In particular, Kac asked the question “can you hear the
shape of a drum?” suppose we know all the frequencies of the modes of
oscillation on some domain Ω. Can we know what is like?
The answer is no, and we can explicitly construct two distinct drums that
sound the same. Fortunately, we get an affirmative answer if we restrict ourselves
a bit. If we require to be convex, and has a real analytic boundary, then yes!
In fact, we have the following result: let
N
(
λ
0
) be the number of eigenvalues
less than λ
0
. Then we can show that
4π
2
lim
λ
0
→∞
N(λ
0
)
λ
0
= Area(Ω).
5 Distributions
5.1 Distributions
When performing separation of variables, we also have the problem of convergence
as well. To perform separation of variables, we first find some particular solutions
of the form, say,
X
(
x
)
Y
(
y
)
Z
(
z
). We know that these solve, say, the wave equation.
However, what we do next is take an infinite sum of these functions. First of
all, how can we be sure that this converges at all? Even if it did, how do we
know that the sum satisfies the differential equation? As we have seen in Fourier
series, an infinite sum of continuous functions can be discontinuous. If it is not
even continuous, how can we say it is a solution of a differential equation?
Hence, at first people were rather skeptical of this method. They thought
that while these methods worked, they only work on a small, restricted domain
of problems. However, later people realized that this method is indeed rather
general, as long as we allow some more “generalized” functions that are not
functions in the usual sense. These are known as distributions.
To define a distribution, we first pick a class of “nice” test functions, where
“nice” means we can do whatever we want to them (e.g. differentiate, integrate
etc.) A main example is the set of infinitely smooth functions with compact
support on some set
K
Ω, written
C
cpt
(Ω) or
D
(Ω). For example, we can
have the bump function defined by
φ(x) =
(
e
1/(1x
2
)
|x| < 1
0 otherwise.
x
We now define a distribution
T
to be a linear map
T
:
D
(Ω)
R
. For those
who are doing IB Linear Algebra, this is the dual space of the space of (“nice”)
real-valued functions. For φ D(Ω), we often write its image under T as T [φ].
Example.
The simplest example is just an ordinary function that is integrable
over any compact region. Then we define the distribution T
f
as
T
f
[φ] =
Z
f(x)φ(x) dx.
Note that this is a linear map since integration is linear (and multiplication
is commutative and distributes over addition). Hence every function “is” a
distribution.
Of course, this would not be interesting if we only had ordinary functions.
The most important example of a distribution (that is not an ordinary function)
is the Dirac delta “function”.
Definition (Dirac-delta). The Dirac-delta is a distribution defined by
δ[φ] = φ(0).
By analogy with the first case, we often abuse notation and write
δ[φ] =
Z
δ(x)φ(x) dx,
and pretend
δ
(
x
) is an actual function. Of course, it is not really a function, i.e.
it is not a map
R
. If it were, then we must have
δ
(
x
) = 0 whenever
x 6
= 0,
since
δ
[
φ
] =
φ
(0) only depends on what happens at 0. But then this integral
will just give 0 if
δ
(0)
R
. Some people like to think of this as a function that
is zero anywhere and “infinitely large” everywhere else. Formally, though, we
have to think of it as a distribution.
Although distributions can be arbitrarily singular and insane, we can nonethe-
less define all their derivatives, defined by
T
0
[φ] = T [φ
0
].
This is motivated by the case of regular functions, where we get, after integrating
by parts,
Z
f
0
(x)φ(x) dx =
Z
f(x)φ
0
(x) dx,
with no boundary terms since we have a compact support. Since
φ
is infinitely
differentiable, we can take arbitrary derivatives of distributions.
So even though distributions can be crazy and singular, everything can be
differentiated. This is good.
Generalized functions can occur as limits of sequences of normal functions.
For example, the family of functions
G
n
(x) =
n
π
exp(n
2
x
2
)
are smooth for any finite n, and G
n
[φ] δ[φ] for any φ.
t
D
It thus makes sense to define
δ
0
(φ) = δ[φ
0
] = φ
0
(0),
as this is the limit of the sequence
lim
n→∞
Z
G
0
n
(x)φ(x) dx.
It is often convenient to think of
δ
(
x
) as
lim
n→∞
G
n
(
x
), and
δ
0
(
x
) =
lim
n→∞
G
0
n
(
x
)
etc, despite the fact that these limits do not exist as functions.
We can look at some properties of δ(x):
Translation:
Z
−∞
δ(x a)φ(x) dx =
Z
−∞
δ(y)φ(y + a) dx.
Scaling:
Z
−∞
δ(cx)φ(x) dx =
Z
−∞
δ(y)φ
y
c
dy
|c|
=
1
|c|
φ(0).
These are both special cases of the following: suppose
f
(
x
) is a continuously
differentiable function with isolated simple zeros at
x
i
. Then near any of
its zeros x
i
, we have f(x) (x x
i
)
f
x
x
i
. Then
Z
−∞
δ(f(x))φ(x) dx =
n
X
i=1
Z
−∞
δ
(x x
i
)
f
x
x
i
!
φ(x) dx
=
n
X
i=1
1
|f
0
(x
i
)|
φ(x
i
).
We can also expand the
δ
-function in a basis of eigenfunctions. Suppose we live
in the interval [L, L], and write a Fourier expansion
δ(x) =
X
nZ
ˆ
δ
n
e
inπx/L
with
ˆ
δ
n
=
1
2L
Z
L
L
e
inπx/L
δ(x) dx =
1
2L
So we have
δ(x) =
1
2L
X
nZ
e
inπx/L
.
This does make sense as a distribution. We let
S
N
δ(x) =
1
2L
N
X
n=N
e
inπx/L
.
Then
lim
N→∞
Z
L
L
S
N
δ(x)φ(x) dx = lim
N→∞
1
2L
Z
L
L
N
X
n=N
e
inπx/L
φ(x) dx
= lim
N→∞
N
X
n=N
"
1
2L
Z
L
L
e
inπx/L
φ(x) dx
#
= lim
N→∞
N
X
n=N
ˆ
φ
n
= lim
N→∞
N
X
n=N
ˆ
φ
n
e
inπ0/L
= φ(0),
since the Fourier series of the smooth function
φ
(
x
) does converge for all
x
[L, L].
We can equally well expand
δ
(
x
) in terms of any other set of orthonormal
eigenfunctions. Let
{y
n
(
x
)
}
be a complete set of eigenfunctions on [
a, b
] that are
orthogonal with respect to a weight function w(x). Then we can write
δ(x ξ) =
X
n
c
n
y
n
(x),
with
c
n
=
Z
b
a
y
n
(x)δ(x ξ)w(x) dx = y
n
(ξ)w(ξ).
So
δ(x ξ) = w(ξ)
X
n
y
n
(ξ)y
n
(x) = w(x)
X
n
y
n
(ξ)y
n
(x),
using the fact that
δ(x ξ) =
w(ξ)
w(x)
δ(x ξ).
5.2 Green’s functions
One of the main uses of the
δ
function is the Green’s function. Suppose we wish
to solve the 2nd order ordinary differential equation
Ly
=
f
, where
f
(
x
) is a
bounded forcing term, and L is a bounded differential operator, say
L = α(x)
2
x
2
+ β(x)
x
+ γ(x).
As a first step, we try to find the Green’s function for
L
, which we shall write as
G(x, ξ), which obeys
LG(x, ξ) = δ(x ξ).
Given G(x, ξ), we can then define
y(x) =
Z
b
a
G(x, ξ)f(ξ) dξ.
Then we have
Ly =
Z
b
a
LG(x, ξ)f(ξ) dξ =
Z
b
a
δ(x ξ)f(ξ) = f(x).
We can formally say
y
=
L
1
f
, and the Green’s function shows that
L
1
means
an integral.
Hence if we can solve for Green’s function, then we have solved the differential
equation, at least in terms of an integral.
To find the Green’s function G(x, ξ) obeying the boundary conditions
G(a, ξ) = G(b, ξ) = 0,
first note that
LG
= 0 for all
x
[
a, ξ
)
(
ξ, b
], i.e. everywhere except
ξ
itself.
Thus we must be able to expand it in a basis of solutions in these two regions.
Suppose
{y
1
(
x
)
, y
2
(
x
)
}
are a basis of solutions to
LG
= 0 everywhere on
[a, b], with boundary conditions y
1
(a) = 0, y
2
(b) = 0. Then we must have
G(x, ξ) =
(
A(ξ)y
1
(x) a x < ξ
B(ξ)y
2
(x) ξ < x b
So we have a whole family of solutions. To fix the coefficients, we must decide
how to join these solutions together over x = ξ.
If
G
(
x, ξ
) were discontinuous at
x
=
ξ
, then
x
G|
x=ξ
would involve a
δ
function, while
2
x
G|
x=ξ
would involve the derivative of the
δ
function. This is
not good, since nothing in
LG
=
δ
(
x ξ
) would balance a
δ
0
. So
G
(
x, ξ
) must
be everywhere continuous. Hence we require
A(ξ)y
1
(ξ) = B(ξ)y
2
(ξ). ()
Now integrate over a small region (ξ ε, ξ + ε) surrounding ξ. Then we have
Z
ξ+ε
ξε
α(x)
d
2
G
dx
2
+ β(x)
dG
dx
+ γ(x)G
dx =
Z
ξ+ε
ξε
δ(x ξ) dx = 1.
By continuity of
G
, we know that the
γG
term does not contribute. While
G
0
is
discontinuous, it is still finite. So the
βG
0
term also does not contribute. So we
have
lim
ε0
Z
ξ+ε
ξε
αG
00
dx = 1.
We now integrate by parts to obtain
lim
ε0
[αG
0
]
ξ+ε
ξε
+
Z
ξ+ε
ξε
α
0
G
0
dx = 1.
Again, by finiteness of G
0
, the integral does not contribute. So we know that
α(ξ)
G
x
ξ
+
G
x
ξ
!
= 1
Hence we obtain
B(ξ)y
0
2
(ξ) A(ξ)y
0
1
(ξ) =
1
α(ξ)
.
Together with (), we know that
A(ξ) =
y
2
(ξ)
α(ξ)W (ξ)
, B(ξ) =
y
1
(ξ)
α(ξ)W (ξ)
,
where W is the Wronskian
W = y
1
y
0
2
y
2
y
0
1
.
Hence, we know that
G(x, ξ) =
1
α(ξ)W (ξ)
(
y
2
(ξ)y
1
(x) a x ξ
y
1
(ξ)y
2
(x) ξ < x b.
Using the step function Θ, we can write this as
G(x, ξ) =
1
α(ξ)W (ξ)
[Θ(ξ x)y
2
(ξ)y
1
(x) + Θ(x ξ)y
1
(ξ)y
2
(x)].
So our general solution is
y(x) =
Z
b
a
G(x, ξ)f(ξ) dξ
=
Z
b
x
f(ξ)
α(ξ)W (ξ)
y
2
(ξ)y
1
(x) dξ +
Z
x
a
f(ξ)
α(ξ)W (ξ)
y
1
(ξ)y
2
(x) dξ.
Example. Consider
Ly = y
00
y = f
for
x
(0
,
1) with
y
(0) =
y
(1) = 0. We choose our basis solution to satisfy
y
00
= y as {sin x, sin(1 x)}. Then we can compute the Wronskian
W (x) = sin x cos(1 x) sin(1 x) cos x = sin 1.
Our Green’s function is
G(x, ξ) =
1
sin 1
[Θ(ξ x) sin(1 ξ) sin x + Θ(x ξ) sin ξ sin(1 x)].
Hence we get
y(x) = sin x
Z
1
x
sin(1 ξ)
sin 1
f(ξ) dξ + sin(1 x)
Z
x
0
sin ξ
sin 1
f(ξ) dξ.
Example.
Suppose we have a string with mass per unit length
µ
(
x
) and held
under a tension T . In the presence of gravity, Newton’s second law gives
µ
2
y
t
2
= T
2
y
x
2
+ µg.
We look for the steady state solution
˙y
= 0 shape of the string, assuming
y(0, t) = y(L, t) = 0. So we get
2
y
x
2
=
µ(x)
T
g.
We look for a Green’s function obeying
2
y
x
2
= δ(x ξ).
This can be interpreted as the contribution of a pointlike mass located at
x
=
ξ
.
0 L
ξ
y
The homogeneous equation y
00
= 0 gives
y = Ax + B(x L).
So we get
G(x, ξ) =
(
A(ξx) 0 x < ξ
B(ξ)(x L) ξ < x L.
Continuity at x = ξ gives
A(ξ)ξ = B(ξ)(ξ L).
The jump condition on the derivative gives
A(ξ) B(ξ) = 1.
We can solve these to get
A(ξ) =
ξ L
L
, B(ξ) =
ξ
L
.
Hence the Green’s function is
G(x, ξ) =
ξ L
L
Θ(ξ x) +
ξ
L
(x L)Θ(x ξ).
Notice that
ξ
is always less that
L
. So the first term has a negative slope; while
ξ is always positive, so the second term has positive slope.
For the general case, we can just substitute this into the integral. We can
give this integral a physical interpretation. We can think that we have many
pointlike particles m
i
at small separations x along the string. So we get
g(x) =
n
X
i=1
G(x, x
i
)
m
i
g
T
,
and in the limit x 0 and n , we get
g(x)
g
T
Z
L
0
G(x, ξ)µ(ξ) dξ.
5.3 Green’s functions for initial value problems
Consider
Ly
=
f
(
t
) subject to
y
(
t
=
t
0
) = 0 and
y
0
(
t
=
t
0
) = 0. So instead of
having two boundaries, we have one boundary and restrict both the value and
the derivative at this boundary.
As before, let
y
1
(
t
)
, y
2
(
t
) be any basis of solutions to
Ly
= 0. The Green’s
function obeys
L(G) = δ(t τ ).
We can write our Green’s function as
G(t, τ) =
(
A(τ)y
1
(t) + B(τ )y
2
(t) t
0
t < τ
C(τ)y
1
(t) + D(τ)y
2
(t) t > τ.
Our initial conditions require that
y
1
(t
0
) y
2
(t
0
)
y
0
1
(t
0
) y
0
2
(t
0
)
A
B
=
0
0
However, we know that the matrix is non-singular (by definition of
y
1
, y
2
being a
basis). So we must have
A, B
= 0 (which makes sense, since
G
= 0 is obviously
a solution for t
0
t < τ ).
Our continuity and jump conditions now require
0 = C(τ)t
1
(t) + D(τ)y
2
(t)
1
α(τ)
= C(τ)y
0
1
(τ) + D(τ )y
0
2
(τ).
We can solve this to get C(τ) and D(τ). Then we get
y(t) =
Z
t
0
G(t, τ)f(τ) dτ =
Z
t
t
0
G(t, τ)f(τ) dτ,
since we know that when
τ > t
, the Green’s function
G
(
t, τ
) = 0. Thus the
solution
y
(
t
) depends on the forcing term term
f
only for times
< t
. This
expresses causality!
Example.
For example, suppose we have
¨y
+
y
=
f
(
t
) with
f
(0) =
˙y
(0) = 0.
Then we have
G(t, τ) = Θ(t τ )[C(τ) cos(t τ ) + D(τ) sin(t τ )]
for some
C
(
τ
)
, D
(
τ
). The continuity and jump conditions gives
D
(
τ
) = 1
, C
(
τ
) =
0. So we get
G(t, τ) = Θ(t τ ) sin t(τ).
So we get
y(t) =
Z
t
0
sin(t τ )f(τ) dτ.
6 Fourier transforms
6.1 The Fourier transform
So far, we have considered writing a function
f
:
S
1
C
(or a periodic function)
as a Fourier sum
f(θ) =
X
nZ
ˆ
f
n
e
inθ
,
ˆ
f
n
=
1
2π
Z
π
π
e
inθ
f(θ) dθ.
What if we don’t have a periodic function, but just an arbitrary function
f : R C? Can we still do Fourier analysis? Yes!
To start with, instead of
ˆ
f
n
=
1
2π
R
π
π
e
inθ
f
(
θ
) d
θ
, we define the Fourier
transform as follows:
Definition (Fourier transform). The Fourier transform of an (absolutely inte-
grable) function f : R C is defined as
˜
f(k) =
Z
−∞
e
ikx
f(x) dx
for all k R. We will also write
˜
f(k) = F[f(x)].
Note that for any k, we have
|
˜
f(k)| =
Z
−∞
e
ikx
f(x) dx
Z
−∞
|e
ikx
f(x)| dx =
Z
−∞
|f(x)| dx.
Since we have assumed that our function is absolutely integrable, this is finite,
and the definition makes sense.
We can look at a few basic properties of the Fourier transform.
(i)
Linearity: if
f, g
:
R C
are absolutely integrable and
c
1
, c
2
are constants,
then
F[c
1
f(x) + c
2
g(x)] = c
1
F[f(x)] + c
2
F[g(x)].
This follows directly from the linearity of the integral.
(ii) Translation:
F[f(x a)] =
Z
R
e
ikx
f(x a) dx
Let y = x a. Then we have
=
Z
R
e
ik(y+a)
f(y) dy
= e
ika
Z
R
e
iky
f(y) dy
= e
ika
F[f(x)],
So a translation in x-space becomes a re-phasing in k-space.
(iii) Re-phasing:
F[e
i`x
f(x)] =
Z
−∞
e
ikx
e
i`x
f(x) dx
=
Z
−∞
e
i(k+`)x
f(x) dx
=
˜
f(k + `),
where ` R and
˜
f(x) = F[f(x)].
(iv) Scaling:
F[f(cx)] =
Z
−∞
e
ikx
f(cx) dx
=
Z
−∞
e
iky/c
f(y)
dy
|c|
=
1
|c|
˜
f
k
c
.
Note that we have to take the absolute value of
c
, since if we replace
x
by
y/c
and
c
is negative, then we will flip the bounds of the integral. The
extra minus sign then turns c into c = |c|.
(v) Convolutions: for functions f, g : R C, we define the convolution as
f g(x) =
Z
−∞
f(x y)g(y) dy.
We then have
F[f g(x)] =
Z
−∞
e
ikx
Z
−∞
f(x y)g(y) dy
dx
=
Z
R
2
e
ik(xy)
f(x y)e
iky
g(y) dy dx
=
Z
R
e
iku
f(u) du
Z
R
e
iky
g(y) dy
= F[f]F[g],
where
u
=
x y
. So the Fourier transform of a convolution is the product
of individual Fourier transforms.
(vi)
The most useful property of the Fourier transform is that it “turns differ-
entiation into multiplication”. Integrating by parts, we have
F[f
0
(x)] =
Z
−∞
e
ikx
df
dx
dx
=
Z
−∞
d
dx
(e
ikx
)f(x) dx
Note that we don’t have any boundary terms since for the function to be
absolutely integrable, it has to decay to zero as we go to infinity. So we
have
= ik
Z
−∞
e
ikx
f(x) dx
= ikF[f(x)]
Conversely,
F[xf(x)] =
Z
−∞
e
ikx
xf(x) dx = i
d
dk
Z
−∞
e
ikx
f(x) dx = i
˜
f
0
(k).
This are useful results, since if we have a complicated function to find the Fourier
transform of, we can use these to break it apart into smaller pieces and hopefully
make our lives easier.
Suppose we have a differential equation
L()y = f,
where
L() =
p
X
r=0
c
r
d
r
dx
r
is a differential operator of pth order with constant coefficients.
Taking the Fourier transform of both sides of the equation, we find
F[L()y] = F[f (x)] =
˜
f(k).
The interesting part is the left hand side, since the Fourier transform turns
differentiation into multiplication. The left hand side becomes
c
0
˜y(k) + c
1
ik˜y(k) + c
2
(ik)
2
˜y(k) + ··· + c
p
(ik)
p
˜y(k) = L(ik)˜y(k).
Here
L
(
ik
) is a polynomial in
ik
. Thus taking the Fourier transform has changed
our ordinary differential equation into the algebraic equation
L(ik)˜y(k) =
˜
f(k).
Since L(ik) is just multiplication by a polynomial, we can immediately get
˜y(k) =
˜
f(k)
L(ik)
.
Example. For φ : R
n
C, suppose we have the equation
2
φ m
2
φ = ρ(x),
where
2
=
n
X
i=1
d
2
dx
2
i
is the Laplacian on R
n
.
We define the n dimensional Fourier transform by
˜
φ(k) =
Z
R
n
e
ik·x
f(x) dx = F[φ(x)]
where now k R
n
is an n-dimensional vector. So we get
F[
2
φ m
2
φ] = F[ρ] = ˜ρ(k).
The first term is
Z
R
n
e
ik·x
· φ d
n
x =
Z
R
n
(e
ik·x
) · φ d
n
x
=
Z
R
n
2
(e
ik·x
)φ d
n
x
= k · k
Z
R
n
e
ik·x
φ(x) d
n
x
= k · k
˜
φ(k).
So our equation becomes
k · k
˜
φ(k) m
2
˜
φ(k) = ˜ρ(k).
So we get
˜
φ(k) =
˜ρ(k)
|k|
2
+ m
2
.
So differential equations are trivial in
k
space. The problem is, of course,
that we want our solution in
x
space, not
k
space. So we need to find a way to
restore our function back to x space.
6.2 The Fourier inversion theorem
We need to be able to express our original function
f
(
x
) in terms of its Fourier
transform
˜
f
(
k
). Recall that in the periodic case where
f
(
x
) =
f
(
x
+
L
), we have
f(x) =
X
nZ
ˆ
f
n
e
2inxπ/L
.
We can try to obtain a similar expression for a non-periodic function
f
:
R C
by taking the limit L . Recall that our coefficients are defined by
ˆ
f
n
=
1
L
Z
L/2
L/2
e
2inπx/L
f(x) dx.
Now define
k =
2π
L
.
So we have
f(x) =
X
nZ
e
inxk
k
2π
Z
L/2
L/2
e
inxk
f(u) du
As we take the limit L , we can approximate
Z
L/2
L/2
e
ixnk
f(x) dx =
Z
−∞
e
ix(nk)
f(x) dx =
˜
f(nk).
Then for a non-periodic function, we have
f(x) = lim
k→∞
X
nZ
k
2π
e
inkx
˜
f(nk) =
1
2π
Z
−∞
e
ikx
˜
f(k) dk.
So
f(x) = F
1
[f(k)] =
1
2π
Z
−∞
e
ikx
˜
f(k) dk.
This is a really dodgy argument. We first took the limit as
L
to turn
the integral from
R
L
L
to
R
−∞
. Then we take the limit as
k
0. However,
we can’t really do this, since
k
is defined to be 2
π/L
, and both limits have
to be taken at the same time. So we should really just take this as a heuristic
argument for why this works, instead of a formal proof.
Nevertheless, note that the inverse Fourier transform looks very similar to
the Fourier transform itself. The only differences are a factor of
1
2π
, and the sign
of the exponent. We can get rid of the first difference by evenly distributing the
factor among the Fourier transform and inverse Fourier transform. For example,
we can instead define
F[f(x)] =
1
2π
Z
−∞
e
ikx
f(x) dx.
Then F
1
looks just like F apart from having the wrong sign.
However, we will stick to our original definition so that we don’t have to deal
with messy square roots. Hence we have the duality
F[f(x)] = F
1
[f(x)](2π).
This is useful because it means we can use our knowledge of Fourier transforms
to compute inverse Fourier transform.
Note that this does not occur in the case of the Fourier series. In the Fourier
series, we obtain the coefficients by evaluating an integral, and restore the
original function by taking a discrete sum. These operations are not symmetric.
Example. Recall that we defined the convolution as
f g(x) =
Z
−∞
f(x y)g(y) dy.
We then computed
F[f g(x)] =
˜
f(k)
˜
f(g).
It follows by definition of the inverse that
F
1
[
˜
f(kg(k)] = f g(x).
Therefore, we know that
F[f(x)g(x)] = 2π
Z
−∞
˜
f(k `)˜g(`) d` = 2π
˜
f ˜g(k).
Example. Last time we saw that if
2
φ m
2
φ = ρ(x), then
˜
φ(k) =
˜ρ(k)
|k|
2
+ m
2
.
Hence we can retrieve our φ as
φ(x) = F
1
[
˜
φ(k)] =
1
(2π)
n
Z
R
n
e
ik·x
˜ρ(k)
|k|
2
+ m
2
d
n
x.
Note that we have (2
π
)
n
instead of 2
π
since we get a factor of 2
π
for each
dimension (and the negative sign was just brought down from the original
expression).
Since we are taking the inverse Fourier transform of a product, we know that
the result is just a convolution of F
1
[˜ρ(k)] = ρ(x) and F
1
1
|k|
2
+m
2
.
Recall that when we worked with Green’s function, if we have a forcing term,
then the final solution is the convolution of our Green’s function with the forcing
term. Here we get a similar expression in higher dimensions.
6.3 Parseval’s theorem for Fourier transforms
Theorem
(Parseval’s theorem (again))
.
Suppose
f, g
:
R C
are sufficiently
well-behaved that
˜
f
and
˜g
exist and we indeed have
F
1
[
˜
f
] =
f, F
1
[
˜g
] =
g
.
Then
(f, g) =
Z
R
f
(x)g(x) dx =
1
2π
(
˜
f, ˜g).
In particular, if f = g, then
kfk
2
=
1
2π
k
˜
fk
2
.
So taking the Fourier transform preserves the
L
2
norm (up to a constant
factor of
1
2π
).
Proof.
(f, g) =
Z
R
f
(x)g(x) dx
=
Z
−∞
f
(x)
1
2π
Z
−∞
e
ikx
˜g(x) dk
dx
=
1
2π
Z
−∞
Z
−∞
f
(x)e
ikx
dx
˜g(k) dk
=
1
2π
Z
−∞
Z
−∞
f(x)e
ikx
dx
˜g(k) dk
=
1
2π
Z
−∞
˜
f
(kg(k) dk
=
1
2π
(
˜
f, ˜g).
6.4 A word of warning
Suppose f(x) is defined by
f(x) =
(
1 |x| < 1
0 |x| 1.
x
y
This function looks rather innocent. Sure it has discontinuities at
±
1, but these
are not too absurd, and we are just integrating. This is certainly absolutely
integrable. We can easily compute the Fourier transform as
˜
f(k) =
Z
−∞
e
ikx
f(x) dx =
Z
1
1
e
ikx
dx =
2 sin k
k
.
Is our original function equal to the integral F
1
[
˜
f(l)]? We have
F
1
2 sin k
k
=
1
π
Z
−∞
e
ikx
sin k
k
dk.
This is hard to do. So let’s first see if this function is absolutely integrable. We
have
Z
−∞
e
ikx
sin k
k
dx =
Z
−∞
sin k
k
dk
= 2
Z
0
sin k
k
dk
2
N
X
n=0
Z
(n+3/4)π
(n+1/4)π
sin k
k
dk.
The idea here is instead of looking at the integral over the whole real line, we
just pick out segments of it. In these small segments, we know that
|sin k|
1
2
.
So we can bound this by
Z
−∞
e
ikx
sin k
k
dx 2
N
X
n=0
1
2
Z
(n+3/4)π
(n+1/4)π
dk
k
2
N
X
n=0
Z
(n+3/4)π
(n+1/4)π
dk
(n + 1)π
=
2
2
N
X
n=0
1
n + 1
,
and this diverges as
N
. So we don’t have absolute integrability. So we
have to be careful when we are doing these things. Well, in theory. We actually
won’t. We’ll just ignore this issue.
6.5 Fourier transformation of distributions
We have
F[δ(x)] =
Z
−∞
e
ikx
δ(x) dx = 1.
Hence we have
F
1
[1] =
1
2π
Z
−∞
e
ikx
dk = δ(x).
Of course, it is extremely hard to make sense of this integral. It quite obviously
diverges as a normal integral, but we can just have faith and believe this makes
sense as long as we are talking about distributions.
Similarly, from our rules of translations, we get
F[δ(x a)] = e
ika
and
F[e
i`x
] = 2πδ(k `),
Hence we get
F[cos(`x)] = F
1
2
(e
i`x
+ e
i`x
)
=
1
2
F[e
i`x
]+
1
2
F[e
i`x
] = π[δ(k`)+δ(k+`)].
We see that highly localized functions in
x
-space have very spread-out behaviour
in k-space, and vice versa.
6.6 Linear systems and response f unctions
Suppose we have an amplifier that modifies an input signal
I
(
t
) to produce an
output
O
(
t
). Typically, amplifiers work by modifying the amplitudes and phases
of specific frequencies in the output. By Fourier’s inversion theorem, we know
I(t) =
1
2π
Z
e
t
˜
I(ω) dω.
This
˜
I(ω) is the resolution of I(t).
We specify what the amplifier does by the transfer function
˜
R
(
ω
) such that
the output is given by
O(t) =
1
2π
Z
−∞
e
t
˜
R(ω)
˜
I(ω) dω.
Since this
˜
R
(
ω
)
˜
I
(
ω
) is a product, on computing
O
(
t
) =
F
1
[
˜
R
(
ω
)
˜
I
(
ω
)], we
obtain a convolution
O(t) =
Z
−∞
R(t u)I(u) du,
where
R(t) =
1
2π
Z
−∞
e
t
˜
R(ω) dω
is the response function. By plugging it directly into the equation above, we see
that R(t) is the response to an input signal I(t) = δ(t).
Now note that causality implies that the amplifier cannot “respond” before
any input has been given. So we must have R(t) = 0 for all t < 0.
Assume that we only start providing input at t = 0. Then
O(t) =
Z
−∞
R(t u)I(u) du =
Z
t
0
R(t u)I(u) du.
This is exactly the same form of solution as we found for initial value problems
with the response function R(t) playing the role of the Green’s function.
6.7 General form of transfer function
To model the situation, suppose the amplifier’s operation is described by the
ordinary differential equation
I(t) = L
m
O(t),
where
L
m
=
m
X
i=0
a
i
d
i
dt
i
with
a
i
C
. In other words, we have an
m
th order ordinary differential equation
with constant coefficients, and the input is the forcing function. Notice that
we are not saying that
O
(
t
) can be expressed as a function of derivatives of
I
(
t
). Instead,
I
(
t
) influences
O
(
t
) by being the forcing term the differential
equation has to satisfy. This makes sense because when we send an input into
the amplifier, this input wave “pushes” the amplifier, and the amplifier is forced
to react in some way to match the input.
Taking the Fourier transform and using F
dO
dt
=
˜
O(ω), we have
˜
I(ω) =
m
X
j=0
a
j
()
j
˜
O(ω).
So we get
˜
O(ω) =
˜
I(ω)
a
0
+ a
1
+ ··· + ()
m
a
m
.
Hence, the transfer function is just
˜
R(ω) =
1
a
0
+ a
1
+ ··· + ()
m
a
m
.
The denominator is an
n
th order polynomial. By the fundamental theorem of
algebra, it has
m
roots, say
c
j
C
for
j
= 1
, ··· , J
, where
c
j
has multiplicity
k
j
. Then we can write our transfer function as
˜
R(ω) =
1
a
m
J
Y
j=1
1
( c
j
)
k
j
.
By repeated use of partial fractions, we can write this as
˜
R(ω) =
J
X
j=1
k
j
X
r=1
Γ
rj
( c
j
)
r
for some constants Γ
rj
C.
By linearity of the (inverse) Fourier transform, we can find
O
(
t
) if we know
the inverse Fourier transform of all functions of the form
1
( α)
p
.
To compute this, there are many possible approaches. We can try to evaluate
the integral directly, learn some fancy tricks from complex analysis, or cheat,
and let the lecturer tell you the answer. Consider the function
h
0
(t) =
(
e
αt
t > 0
0 otherwise
The Fourier transform is then
˜
h
o
(ω) =
Z
−∞
e
t
h
0
(t) dt =
Z
0
e
(α)t
dt =
1
α
provided Re(α) < 0 (so that e
(α)t
0 as t ). Similarly, we can let
h
1
(t) =
(
te
αt
t > 0
0 otherwise
Then we get
˜
h
1
(ω) = F[th
0
(t)] = i
d
dω
F[h
0
(t)] =
1
( α)
2
.
Proceeding inductively, we can define
h
p
(t) =
(
t
p
p!
e
αt
t > 0
0 otherwise
,
and this has Fourier transform
˜
h
p
(ω) =
1
( α)
p+1
,
again provided
Re
(
α
)
<
0. So the response function is a linear combination
of these functions
h
p
(
t
) (if any of the roots
c
j
have non-negative real part,
then it turns out the system is unstable, and is better analysed by the Laplace
transform). We see that the response function does indeed vanish for all
t <
0.
In fact, each term (except
h
0
) increases from zero at
t
= 0 to rise to some
maximum before eventually decaying exponentially.
Fourier transforms can also be used to solve ordinary differential equations
on the whole real line
R
, provided the functions are sufficiently well-behaved.
For example, suppose y : R R solves
y
00
A
2
y = f(x)
with y and y
0
0 as |x| . Taking the Fourier transform, we get
˜y(k) =
˜
f(k)
k
2
+ A
2
.
Since this is a product, the inverse Fourier transform will be a convolution of
f
(
x
)
and an object whose Fourier transform is
1
k
2
+A
2
. Again, we’ll cheat. Consider
h(x) =
1
2µ
e
µ|x|
.
with µ > 0. Then
˜
f(k) =
1
2µ
Z
−∞
e
ikx
e
µ|x|
dx
=
1
µ
Re
Z
0
e
(µ+ik)x
dx
=
1
µ
Re
1
ik + µ
=
1
µ
2
+ k
2
.
Therefore we get
y(x) =
1
2A
Z
−∞
e
A|xu|
f(u) du.
So far, we have done Fourier analysis over some abelian groups. For example,
we’ve done it over
S
1
, which is an abelian group under multiplication, and
R
,
which is an abelian group under addition. We will next look at Fourier analysis
over another abelian group, Z
m
, known as the discrete Fourier transform.
6.8 The discrete Fourier transform
Recall that the Fourier transform is defined as
˜
f(k) =
Z
R
e
ikx
f(x) dx.
To find the Fourier transform, we have to know how to perform the integral. If
we cannot do the integral, then we cannot do the Fourier transform. However,
it is usually difficult to perform integrals for even slightly more complicated
functions.
A more serious problem is that we usually don’t even have a closed form of
the function
f
for us to perform the integral. In real life,
f
might be some radio
signal, and we are just given some data of what value
f
takes at different points..
There is no hope that we can do the integral exactly.
Hence, we first give ourselves a simplifying assumption. We suppose that
f
is mostly concentrated in some finite region [
R, S
]. More precisely,
|f
(
x
)
|
1
for x 6∈ [R, S] for some R, S > 0. Then we can approximate our integral as
˜
f(k) =
Z
S
R
e
ikx
f(x) dx.
Afterwards, we perform the integral numerically. Suppose we “sample”
f
(
x
) at
x = x
j
= R + j
R + S
N
for N a large integer and j = 0, 1, ··· , N 1. Then
˜
f(k)
R + S
N
N1
X
j=0
f(x
j
)e
ikx
j
.
This is just the Riemann sum.
Similarly, our computer can only store the result
˜
f
(
k
) for some finite list of
k. Let’s choose these to be at
k = k
m
=
2πm
R + S
.
Then after some cancellation,
˜
f(k
m
)
R + S
N
e
2πimR
R+S
N1
X
j=0
f(x
j
)e
2πi
N
jm
= (R + S)e
2πimR
R+S
1
N
N1
X
j=0
f(x
j
)ω
jm
,
where
ω = e
2πi
N
is an N th root of unity. We call the thing in the brackets
F (m) =
1
N
N1
X
j=0
f(x
j
)ω
jm
.
Of course, we’ve thrown away lots of information about our original function
f
(
x
), since we made approximations all over the place. For example, we have
already lost all knowledge of structures varying more rapidly than our sampling
interval
R+S
N
. Also,
F
(
m
+
N
) =
F
(
m
), since
ω
N
= 1. So we have “forced” some
periodicity into our function F , while
˜
f(k) itself was not necessarily periodic.
For the usual Fourier transform, we were able to re-construct our original
function from the
˜
f
(
k
), but here we clearly cannot. However, if we know the
F
(
m
) for all
m
= 0
,
1
, ··· , N
1, then we can reconstruct the exact values of
f
(
x
j
) for all
j
by just solving linear equations. To make this more precise, we
want to put what we’re doing into the linear algebra framework we’ve been using.
Let
G
=
{
1
, ω, ω
2
, ··· , ω
N1
}
. For our purposes below, we can just treat this as
a discrete set, and
ω
i
are just meaningless symbols that happen to visually look
like the powers of our Nth roots of unity.
Consider a function g : G C defined by
g(ω
j
) = f(x
j
).
This is actually nothing but just a new notation for
f
(
x
j
). Then using this new
notation, we have
F (m) =
1
N
N1
X
j=0
ω
jm
g(ω
j
).
The space of all functions
g
:
G C
is a finite-dimensional vector space,
isomorphic to C
N
. This has an inner product
(f, g) =
1
N
N1
X
j=0
f
(ω
j
)g(ω
j
).
This inner product obeys
(g, f ) = (f, g)
(f, c
1
g
1
+ c
2
g
2
) = c
1
(f, g
1
) + c
2
(f, g
2
)
(f, f) 0 with equality iff f(ω
j
) = 0
Now let e
n
: G C be the function
e
m
(ω
j
) = ω
jm
.
Then the set of functions
{e
m
}
for
m
= 0
, ··· , N
1 is an orthonormal basis
with respect to our inner product. To show this, we can compute
(e
m
, e
m
) =
1
N
N1
X
j=0
e
m
(ω
j
)e
m
(ω
j
)
=
1
N
N1
X
j=0
ω
jm
ω
jm
=
1
N
N1
X
j=0
1
= 1.
For n 6= m, we have
(e
n
, e
m
) =
1
N
N1
X
j=0
e
n
(ω
j
)e
m
(ω
j
)
=
1
N
N1
X
j=0
ω
j(mn)
=
1
N
1 ω
(mn)N
1 ω
mn
.
Since
mn
is an integer and
ω
is an
N
th root of unity, we know that
ω
(mn)N
= 1.
So the numerator is 0. However, since
n 6
=
m
,
m n 6
= 0. So the denominator is
non-zero. So we get 0. Hence
(e
n
, e
m
) = 0.
We can now rewrite our F (m) as
F (m) =
1
N
N1
X
j=0
ω
jm
f(x
j
) =
1
N
N1
X
j=0
e
m
(ω
j
)g(ω
j
) = (e
m
, g).
Hence we can expand our g as
g =
N1
X
m=0
(e
m
, g)e
m
=
N1
X
m=0
F (m)e
m
.
Writing f instead of g, we recover the formula
f(x
j
) = g(ω
j
) =
N1
X
m=0
F (m)e
m
(ω
j
).
If we forget about our
f
s and just look at the
g
, what we have effectively done is
take the Fourier transform of functions taking values on
G
=
{
1
, ω, ··· , ω
N1
}
=
Z
N
.
This is exactly analogous to what we did for the Fourier transform, except
that everything is now discrete, and we don’t have to worry about convergence
since these are finite sums.
6.9 The fast Fourier transform*
What we’ve said so far is we’ve defined
F (m) =
1
N
N1
X
j=0
ω
mj
g(ω
j
).
To compute this directly, even given the values of
ω
mj
for all
j, m
, this takes
N
1 additions and
N
+ 1 multiplications. This is 2
N
operations for a single
value of m. To know F (m) for all m, this takes 2N
2
operations.
This is a problem. Historically, during the cold war, especially after the
Cuban missile crisis, people were in fear that the world will one day go into
a huge nuclear war, and something had to be done to prevent this. Thus the
countries decided to come up with a treaty to ensure people don’t perform
nuclear testings anymore.
However, the obvious problem is that we can’t easily check if other countries
are doing nuclear tests. Nuclear tests are easy to spot if the test is done above
ground, since there will be a huge explosion. However, if it is done underground,
it is much harder to test.
They then came up with the idea to put some seismometers all around
the world, and record vibrations in the ground. To distinguish normal crust
movements from nuclear tests, they wanted to take the Fourier transform and
analyze the frequencies of the vibrations. However, they had a large amount of
data, and the value of
N
is on the order of magnitude of a few million. So 2
N
2
will be a huge number that the computers at that time were not able to handle.
So they need to find a trick to perform Fourier transforms quickly. This is known
as the fast Fourier transform. Note that this is nothing new mathematically.
This is entirely a computational trick.
Now suppose N = 2M . We can write
F (m) =
1
2M
2M1
X
j=0
ω
jm
g(ω
j
)
=
1
2
"
1
M
M1
X
k=0
ω
2km
g(ω
2k
) + ω
(2k+1)m
g(ω
2k+1
)
#
.
We now let η be an M th root of unity, and define
G(η
k
) = g(ω
2k
), H(η
k
) = g(ω
2k+1
).
We then have
F (m) =
1
2
"
1
M
M1
X
k=0
η
km
G(η
k
) +
ω
m
M
M1
X
k=0
η
km
H(η
k
)
#
=
1
2
[
˜
G(m) + ω
m
˜
H(m)].
Suppose we are given the values of
˜
G
(
m
) and
˜
H
(
m
) and
ω
m
for all
m
=
{
0
, ··· , N
1
}
. Then we can compute
F
(
m
) using 3 operations per value of
m
.
So this takes 3 ×N = 6M operations for all M.
We can compute
ω
m
for all
m
using at most 2
N
operations, and suppose it
takes
P
M
operations to compute
˜
G
(
m
) for all
m
. Then the number of operations
needed to compute F (m) is
P
2M
= 2P
M
+ 6M + 2M.
Now let N = 2
n
. Then by iterating this, we can find
P
N
4N log
2
N N
2
.
So by breaking the Fourier transform apart, we are able to greatly reduce the
number of operations needed to compute the Fourier transform. This is used
nowadays everywhere for everything.
7 More partial differential equations
7.1 Well-posedness
Recall that to find a unique solution of Laplace’s equation
2
φ
= 0 on a bounded
domain
R
n
, we imposed the Dirichlet boundary condition
φ|
=
f
or the
Neumann boundary condition n · φ|
= g.
For the heat equation
φ
t
=
κ
2
φ
on Ω
×
[0
,
), we asked for
|
×[0,)
=
f
and also φ|
×{0}
= g.
For the wave equation
2
φ
t
2
=
c
2
2
φ
on
×
[0
,
), we imposed
φ|
×[0,)
=
f
,
φ|
×{0}
= g and also
t
φ|
×{0}
= h.
All the boundary and initial conditions restrict the value of
φ
on some co-
dimension 1 surface, i.e. a surface whose dimension is one less than the domain.
This is known as Cauchy data for our partial differential equation.
Definition
(Well-posed problem)
.
A partial differential equation problem is
said to be well-posed if its Cauchy data means
(i) A solution exists;
(ii) The solution is unique;
(iii)
A “small change” in the Cauchy data leads to a “small change” in the
solution.
To understand the last condition, we need to make it clear what we mean by
“small change”. To do this properly, we need to impose some topology on our
space of functions, which is some technicalities we will not go into. Instead, we
can look at a simple example.
Suppose we have the heat equation
t
φ
=
κ
2
φ
. We know that whatever
starting condition we have, the solution quickly smooths out with time. Any
spikiness of the initial conditions get exponentially suppressed. Hence this is
a well-posed problem changing the initial condition slightly will result in a
similar solution.
However, if we take the heat equation but run it backwards in time, we get
a non-well-posed problem. If we provide a tiny, small change in the “ending
condition”, as we go back in time, this perturbation grows exponentially, and
the result could vary wildly.
Another example is as follows: consider the Laplace’s equation
2
φ
on the
upper half plane (x, y) R ×R
0
subject to the boundary conditions
φ(x, 0) = 0,
y
φ(x, 0) = g(x).
If we take g(x) = 0, then φ(x, y) = 0 is the unique solution, obviously.
However, if we take g(x) =
sin Ax
A
, then we get the unique solution
φ(x, y) =
sin(Ax) sinh(Ay)
A
2
.
So far so good. However, now consider the limit as A . Then
g(x) =
sin(Ax)
A
0.
for all x R. However, at the special point φ
π
2A
, y
, we get
φ
π
2A
, y
=
sinh(Ay)
A
2
A
2
e
Ay
,
which is unbounded. So as we take the limit as our boundary conditions
g
(
x
)
0,
we get an unbounded solution.
How can we make sure this doesn’t happen? We’re first going to look at first
order equations.
7.2 Method of characteristics
Curves in the plane
Suppose we have a curve on the plane x : R R
2
given by s 7→ (x(s), y(s)).
Definition
(Tangent vector)
.
The tangent vector to a smooth curve
C
given
by x : R R
2
with x(s) = (x(s), y(s)) is
dx
ds
,
dy
ds
.
If we have some quantity
φ
:
R
2
R
, then its value along the curve
C
is just
φ(x(s), y(s)).
A vector field
V
(
x, y
) :
R
2
R
2
defines a family of curves. To imagine this,
suppose we are living in a river. The vector field tells us the how the water flows
at each point. We can then obtain a curve by starting a point and flow along
with the water. More precisely,
Definition
(Integral curve)
.
Let
V
(
x, y
) :
R
2
R
2
be a vector field. The
integral curves associated to
V
are curves whose tangent
dx
ds
,
dy
ds
is just
V
(
x, y
).
For sufficiently regular vector fields, we can fill the space with different curves.
We will parametrize which curve we are on by the parameter t. More precisely,
we have a curve
B
= (
x
(
t
)
, y
(
t
)) that is transverse (i.e. nowhere parallel) to our
family of curves, and we can label the members of our family by the value of
t
at which they intersect B.
C
4
C
3
C
2
C
1
B(t)
We can thus label our family of curves by (x(s, t), y(s, t)). If the Jacobian
J =
x
s
y
t
x
t
y
s
6= 0,
then we can invert this to find (
s, t
) as a function of (
x, y
), i.e. at any point, we
know which curve we are on, and how far along the curve we are.
This means we now have a new coordinate system (
s, t
) for our points in
R
2
. It turns out by picking the right vector field
V
, we can make differential
equations much easier to solve in this new coordinate system.
The method of characteristics
Suppose φ : R
2
R obeys
a(x, y)
φ
x
+ b(x, y)
φ
y
= 0.
We can define our vector field as
V(x, y) =
a(x, y)
b(x, y)
.
Then we can write the differential equation as
V ·φ = 0.
Along any particular integral curve of V, we have
φ
s
=
dx(s)
ds
φ
x
+
dy(s)
ds
φ
y
= V ·φ,
where the integral curves of V are determined by
x
s
t
= a(x, y),
y
s
t
= b(x, y).
This is known as the characteristic equation.
Hence our partial differential equation just becomes the equation
φ
s
t
= 0.
To get ourselves a well-posed problem, we have to specify our boundary data
along a transverse curve
B
. We pick our transverse curve as
s
= 0, and we
suppose we are given the initial data
φ(0, t) = h(t).
Since φ does not vary with s, our solution automatically is
φ(s, t) = h(t).
Then φ(x, y) is given by inverting the characteristic equations to find t(x, y).
We do a few examples to make this clear.
Example (Trivial). Suppose φ obeys
φ
x
y
= 0.
We are given the boundary condition
φ(0, y) = f(y).
The solution is obvious, since the differential equation tells us the function is
just a function of
y
, and then the solution is obviously
φ
(
x, y
) =
h
(
y
). However,
we can try to use the method of characteristics to practice our skills. Our vector
field is given by
V =
1
0
=
dx
ds
dy
ds
.
Hence, we get
dx
ds
= 1,
dy
ds
= 0.
So we have
x = s + c, y = d.
Our initial curve is given by
y
=
t, x
= 0. Since we want this to be the curve
s = 0, we find our integral curves to be
x = s, y = t.
Now for each fixed t, we have to solve
φ
s
t
= 0, φ(s, t) = h(t) = f (t).
So we know that
φ(x, y) = f(y).
Example. Consider the equation
e
x
φ
x
+
φ
y
= 0.
with the boundary condition
φ(x, 0) = cosh x.
We find that the vector field is
V =
e
x
1
This gives
dx
ds
= e
x
,
dy
ds
= 1.
Thus we have
e
x
= s + c, y = s + d.
We pick our curve as x = t, y = 0. So our relation becomes
e
x
= s + e
t
, y = s.
We thus have
φ
s
t
= 0, φ(s, t) = φ(0, t) = cosh(t) = cosh[ln(y + e
x
)]
So done.
Example.
Let
φ
:
R
2
R
solve the inhomogeneous partial differential equation
x
φ + 2
y
φ = ye
x
with φ(x, x) = sin x.
We can still use the method of characteristics. We have
u =
1
2
.
So the characteristic curves obey
dx
ds
= 1,
dy
ds
= 2.
This gives
x = s + t, y = 2s + t
so that x = y = t at s = 0. We can invert this to obtain the relations
s = y x, t = 2x y.
The partial differential equation now becomes
dφ
ds
t
= u · φ = ye
x
= (2s + t)e
s+t
Note that this is just an ordinary differential equation in
s
because
t
is held
constant. We have the boundary conditions
φ(s = 0, t) = sin t.
So we get
φ(x(s, t), y(s, t)) = (2 t)e
t
(1 e
s
) + sin t + 2se
s+t
.
Putting it in terms of x and y, we have
φ(x, y) = (2 2x + y)e
2xy
+ sin(2x y) + (y 2)e
x
.
We see that our Cauchy data
φ
(
x, x
) =
sin x
should be specified on a curve
B that intersects each characteristic curve exactly once.
If we tried to use a characteristic curve
B
that intersects the characteristics
multiple times, if we want a solution at all, we cannot specify data freely along
B
. its values at the points of intersection have to compatible with the partial
differential equation.
For example, if we have a homogeneous equation, we saw the solution will
be constant along the same characteristic. Hence if our Cauchy data requires
the solution to take different values on the same characteristic, we will have no
solution.
Moreover, even if it is compatible, if
B
does not intersect all characteristics,
we will not get a unique solution. So the solution is not fixed along such
characteristics.
On the other hand, if
B
intersects each characteristic curve transversely,
the problem is well-posed. So there exists a unique solution, at least in the
neighbourhood of
B
. Notice that data is never transferred between characteristic
curves.
7.3 Characteristics for 2nd order partial differential equa-
tions
Whether the method of characteristics works for a 2nd order partial differential
equation, or even higher order ones, depends on the “type” of the differential
equation.
To classify these differential equations, we need to define
Definition
(Symbol and principal part)
.
Let
L
be the general 2nd order differ-
ential operator on R
n
. We can write it as
L =
n
X
i,j=1
a
ij
(x)
2
x
i
x
j
+
n
X
i=1
b
i
(x)
x
i
+ c(X),
where a
ij
(x), b
i
(x), c(x) R and a
ij
= a
ji
(wlog).
We define the symbol σ(k, x) of L to be
σ(k, x) =
n
X
i,j=1
a
ij
(x)k
i
k
j
+
n
X
i=1
b
i
(x)k
i
+ c(x).
So we just replace the derivatives by the variable k.
The principal part of the symbol is the leading term
σ
p
(k, x) =
n
X
i,j=1
a
ij
(x)k
i
k
j
.
Example. If L =
2
, then
σ(k, x) = σ
p
(k, x) =
n
X
i=1
(k
i
)
2
.
If
L
is the heat operator (where we think of the last coordinate
x
n
as the time,
and others as space), then the operator is given by
L =
x
n
n1
X
i=1
2
x
i2
.
The symbol is then
σ(k, x) = k
n
n1
X
i=1
(k
i
)
2
,
and the principal part is
σ
p
(k, x) =
n1
X
i=1
(k
i
)
2
.
Note that the symbol is closely related to the Fourier transform of the
differential operator, since both turn differentiation into multiplication. Indeed,
they are equal if the coefficients are constant. However, we define this symbol
for arbitrary differential operators with non-constant coefficients.
In general, for each x, we can write
σ
p
(k, x) = k
T
A(x)k,
where
A
(
x
) has elements
a
ij
(
x
). Recall that a real symmetric matrix (such as
A) has all real eigenvalues. So we define the following:
Definition
(Elliptic, hyperbolic, ultra-hyperbolic and parabolic differential
operators). Let L be a differential operator. We say L is
elliptic at x if all eigenvalues of A(x) have the same sign. Equivalently, if
σ
p
( ·, x) is a definite quadratic form;
hyperbolic at x if all but one eigenvalues of A(x) have the same sign;
ultra-hyperbolic at x if A(x) has more than one eigenvalues of each sign;
parabolic at x if A(x) has a zero eigenvalue, i.e. σ
p
( ·, x) is degenerate.
We say L is elliptic if L is elliptic at all x, and similarly for the other terms.
These names are supposed to remind us of conic sections, and indeed if we
think of k
T
Ak as an equation in k, then we get a conic section.
Example. Let
L = a(x, y)
2
x
2
+ 2b(x, y)
2
x∂y
+ c(x, y)
2
y
2
+ d(x, y)
x
+ e(x, y)
y
+ f(x, y).
Then the principal part of the symbol is
σ
p
(k, x) =
k
x
k
y
a(x, y) b(x, y)
b(x, y) c(x, y)
k
x
k
y
Then
L
is elliptic at
x
if
b
2
ac <
0; hyperbolic if
b
2
ac >
0; and parabolic if
b
2
ac = 0.
Note that since we only have two dimensions and hence only two eigenvalues,
we cannot possibly have an ultra-hyperbolic equation.
Characteristic surfaces
Definition (Characteristic surface). Given a differential operator L, let
f(x
1
, x
2
, ··· , x
n
) = 0
define a surface C R
n
. We say C is characteristic if
n
X
i,j=1
a
ij
(x)
f
x
i
f
x
j
= (f)
T
A(f) = σ
p
(f, x) = 0.
In the case where we only have two dimensions, a characteristic surface is just a
curve.
We see that the characteristic equation restricts what values
f
can take.
Recall that
f
is the normal to the surface
C
. So in general, at any point, we
can find what the normal of the surface should be, and stitch these together to
form the full characteristic surfaces.
For an elliptic operator, all the eigenvalues of
A
have the same sign. So there
are no non-trivial real solutions to this equation (
f
)
T
A
(
f
) = 0. Consequently,
elliptic operators have no real characteristics. So the method of characteristics
would not be of any use when studying, say, Laplace’s equation, at least if we
want to stay in the realm of real numbers.
If
L
is parabolic, we for simplicity assume that
A
has exactly one zero
eigenvector, say
n
, and the other eigenvalues have the same sign. This is the
case when, say, there are just two dimensions. So we have
An
=
n
T
A
=
0
. We
normalize n such that n · n = 1.
For any f, we can always decompose it as
f = n(n · f) + [f n · (n · f)].
This is a relation that is trivially true, since we just add and subtract the same
thing. Note, however, that the first term points along
n
, while the latter term is
orthogonal to n. To save some writing, we adopt the notation
f = f n(n · f ).
So we have
f = n(n · f) + f
.
Then we can compute
(f)
T
A(f) = [n(n · f ) +
f]
T
A[n(n · f ) +
f]
= (
f)
T
A(
f).
Then by assumption, (
f
)
T
A
(
f
) is definite. So just as in the elliptic case,
there are no non-trivial solutions. Hence, if
f
defines a characteristic surface,
then
f
= 0. In other words,
f
is parallel to
n
. So at any point, the normal to
a characteristic surface must be
n
, and there is only one possible characteristic.
If
L
is hyperbolic, we assume all but one eigenvalues are positive, and let
λ
be the unique negative eigenvalue. We let
n
be the corresponding unit
eigenvector, where we normalize it such that
n ·n
= 1. We say
f
is characteristic
if
0 = (f)
T
A(f)
= [n(n · f ) +
f]
T
A[n(n · f ) +
f]
= λ(n · f )
2
+ (
f)
T
A(
f).
Consequently, for this to be a characteristic, we need
n · f = ±
r
(
f)
T
A(
f)
λ
.
So there are two choices for
n · f
, given any
f
. So hyperbolic equations
have two characteristic surfaces through any point.
This is not too helpful in general. However, in the case where we have two
dimensions, we can find the characteristic curves explicitly. Suppose our curve
is given by
f
(
x, y
) = 0. We can write
y
=
y
(
x
). Then since
f
is constant along
each characteristic, by the chain rule, we know
0 =
f
x
+
f
y
dy
dx
.
Hence, we can compute
dy
dx
=
b ±
b
2
ac
a
.
We now see explicitly how the type of the differential equation influences the
number of characteristics — if
b
2
ac >
0, then we obtain two distinct differential
equations and obtain two solutions; if
b
2
ac
= 0, then we only have one equation;
if b
2
ac < 0, then there are no real characteristics.
Example. Consider
2
y
φ xy
2
x
φ = 0
on
R
2
. Then
a
=
xy, b
= 0
, c
= 1. So
b
2
ac
=
xy
. So the type is elliptic if
xy < 0, hyperbolic if xy > 0, and parabolic if xy = 0.
In the regions where it is hyperbolic, we find
b ±
b
2
ac
a
= ±
1
xy
.
Hence the two characteristics are given by
dy
dx
= ±
1
xy
.
This has a solution
1
3
y
3/2
± x
1/2
= c.
We now let
u =
1
3
y
3/2
+ x
1/2
,
v =
1
3
y
3/2
x
1/2
.
Then the equation becomes
2
φ
u∂v
+ lower order terms = 0.
Example. Consider the wave equation
2
φ
t
2
c
2
2
φ
x
2
= 0
on
R
1,1
. Then the equation is hyperbolic everywhere, and the characteristic
curves are
x ± ct
= const. Let’s look for a solution to the wave equation that
obeys
φ(x, 0) = f(x),
t
φ(x, 0) = g(x).
Now put u = x ct, v = x + ct. Then the wave equation becomes
2
u∂v
= 0.
So the general solution to this is
φ(x, t) = G(u) + H(v) = G(x ct) + H(x + ct).
The initial conditions now fix these functions
f(x) = G(x) + H(x), g(x) = cG
0
(x) + cH
0
(x).
Solving these, we find
φ(x, t) =
1
2
[f(x ct) + f(x + ct)] +
1
2c
Z
x+ct
xct
g(y) dy.
This is d’Alembert’s solution to the 1 + 1 dimensional wave equation.
Note that the value of
φ
at any point (
x, t
) is completely determined by
f, g
in the interval [
x ct, x
+
ct
]. This is known as the (past) domain of dependence
of the solution at (
x, t
), written
D
(
x, t
). Similarly, at any time
t
, the initial
data at (
x
0
,
0) only affects the solution within the region
x
0
ct x x
0
+
ct
.
This is the range of influence of data at x
0
, written D
+
(x
0
).
p
D
(p)
B
S
D
+
(S)
We see that disturbances in the wave equation propagate with speed c.
7.4 Green’s functions for PDEs on R
n
Green’s functions for the heat equation
Suppose φ : R
n
× [0, ) R solves the heat equation
t
φ = D
2
φ,
where D is the diffusion constant, subject to
φ|
R
n
×{0}
= f.
To solve this, we take the Fourier transform of the heat equation in the spatial
variables. For simplicity of notation, we take
n
= 1. Writing
F
[
φ
(
x, t
)] =
˜
φ
(
k, t
),
we get
t
˜
φ(k, t) = Dk
2
˜
φ(k, t)
with the boundary conditions
˜
φ(k, 0) =
˜
f(k).
Note that this is just an ordinary differential equation in
t
, and
k
is a fixed
parameter for each k. So we have
˜
φ(k, t) =
˜
f(k)e
Dk
2
t
.
We now take the inverse Fourier transform and get
φ(x, t) =
1
2π
Z
R
e
ikx
h
˜
f(k)e
Dk
2
t
i
dk
This is the inverse Fourier transform of a product. This thus gives the convolution
φ(x, t) = f F
1
[e
Dk
2
t
].
So we are done if we can find the inverse Fourier transform of the Gaussian.
This is an easy exercise on example sheet 3, where we find
F[e
a
2
x
2
] =
π
a
e
k
2
/4a
2
.
So setting
a
2
=
1
4Dt
,
So we get
F
1
[e
Dk
2
t
] =
1
4πDt
exp
x
2
4Dt
We shall call this
S
1
(
x, t
), where the subscript 1 tells us we are in 1+1 dimensions.
This is known as the fundamental solution of the heat equation. We then get
φ(x, t) =
Z
−∞
f(y)S
1
(x y, t) dy
Example. Suppose our initial data is
f(x) = φ
0
δ(x).
So we start with a really cold room, with a huge spike in temperature at the
middle of the room. Then we get
φ(x, t) =
φ
0
4πDt
exp
x
2
4Dt
.
What this shows, as we’ve seen before, is that if we start with a delta function,
then as time evolves, we get a Gaussian that gets shorter and broader.
Now note that if we start with a delta function, at
t
= 0, everywhere outside
the origin is non-zero. However, after any infinitely small time
t
,
φ
becomes
non-zero everywhere, instantly. Unlike the wave equation, information travels
instantly to all of space, i.e. heat propagates arbitrarily fast according to this
equation (of course in reality, it doesn’t). This fundamentally, is because the
heat equation is parabolic, and only has one characteristic.
Now suppose instead φ satisfies the inhomogeneous, forced, heat equation
t
φ D
2
φ = F (x, t),
but with homogeneous initial conditions
φ|
t=0
= 0. Physically, this represents
having an external heat source somewhere, starting at zero.
Note that if we can solve this, then we have completely solved the heat
equation. If we have an inhomogeneous equation and an inhomogeneous initial
condition, then we can solve this forced problem with homogeneous boundary
conditions to get
φ
F
; and solve the unforced equation with homogeneous equation
to get φ
H
. Then the sum φ = φ
F
+ φ
H
solves the full equation.
As before, we take the Fourier transform of the forced equation with respect
to the spacial variables. As before, we will just do it in the cases with one spacial
dimension. We find
t
˜
φ(k, t) + Dk
2
˜
φ(k, t) =
˜
F (k, t),
with the initial condition
˜
φ(k, 0) = 0.
As before, we have reduced this to a first-order ordinary differential equation in
t. Using an integrating factor, we can rewrite this as
t
[e
Dk
2
t
˜
φ(k, t)] = e
Dk
2
t
˜
F (k, t).
The solution is them
˜
φ(k, t) = e
Dk
2
t
Z
t
0
e
Dk
2
u
˜
F (k, u) du.
We define the Green’s function G(x, t; y, τ )) to be the solution to
[
t
D
2
x
]G(x, t; y, τ ) = δ(x y)δ(t τ).
So the Fourier transform with respect to x gives
˜
G(k, t, y, τ) = e
Dk
2
t
Z
t
0
e
Dk
2
u
e
iky
δ(t τ ) du,
where e
iky
is just the Fourier transform of δ(x y). This is equal to
˜
G(k, t; y, τ) =
(
0 t < τ
e
iky
e
Dk
2
(tτ)
t > τ
= Θ(t τ )e
iky
e
Dk
2
(tτ)
.
Reverting the Fourier transform, we get
G(x, t; y, τ ) =
Θ(t τ )
2π
Z
R
e
ik(xy)
e
Dk
2
(tτ)
dx
This integral is just the inverse Fourier transform of the Gaussian with a phase
shift. So we end up with
G(x, t; y, τ ) =
Θ(t τ )
p
4πD(t τ)
exp
(x y)
2
4D(t τ)
= Θ(t τ )S
1
(x y; t τ).
The solution we seek is then
φ(x, t) =
Z
t
0
Z
R
F (y, τ )G(x, t; y, τ ) dy dτ.
It is interesting that the solution to the forced equation involves the same function
S
1
(
x, t
) as the homogeneous equation with inhomogeneous boundary conditions.
In general, S
n
(x, t) solves
S
n
t
D
2
S
n
= 0
with boundary conditions S
n
(x, 0) = δ
(n)
(x y), and we can find
S
n
(x, t) =
1
(4πDt)
n/2
exp
|x y|
2
4Dt
.
Then in general, given an initial condition φ|
t=0
= f(x), the solution is
φ(x, t) =
Z
f(y)S(x y, t) d
n
y.
Similarly, G
n
(x, t; y, t) solves
G
n
t
D
2
G
n
= δ(t τ )δ
(n)
(x y),
with the boundary condition G
n
(x, 0; y, τ ) = 0. The solution is
G(x, t; y, τ ) = Θ(t τ)S
n
(x y, t τ).
Given our Green’s function, the general solution to the forced heat equation
φ
t
D
2
φ = F (x, t), φ(x, 0) = 0
is just
φ(x, t) =
Z
0
Z
R
n
F (y, τ )G(x, t; y, τ ) d
n
y dτ
=
Z
t
0
Z
R
n
F (y, τ )G(x, t; y, τ ) d
n
y dτ.
Duhamel noticed that we can write this as
φ(x, t) =
Z
t
0
φ
F
(x, t; τ) dτ,
where
φ
F
=
Z
R
F (y, t)S
n
(x y, t τ) d
n
y
solves the homogeneous heat equation with φ
F
|
t=τ
= F (x, τ).
Hence in general, we can think of the forcing term as providing a whole
sequence of “initial conditions” for all
t >
0. We then integrate over the times
at which these conditions were imposed to find the full solution to the forced
problem. This interpretation is called Duhamel’s principle.
Green’s functions for the wave equation
Suppose φ : R
n
× [0, ) C solves the inhomogeneous wave equation
2
φ
t
2
c
2
2
φ = F (x, t)
with
φ(x, 0) =
t
φ(x, 0) = 0.
We look for a Green’s function G
n
(x, t; y, τ ) that solves
G
n
t
c
2
2
G
n
= δ(t τ )δ
(n)
(x y) ()
with the same initial conditions
G
n
(x, 0, y, τ ) =
t
G
n
(x, 0, y, τ ) = 0.
Just as before, we take the Fourier transform of this equation with respect the
spacial variables x. We get
t
˜
G
n
+ c
2
|k|
2
˜
G
n
= δ(t τ )e
ik·y
.
where
˜
G
n
=
˜
G
n
(k, t, y, τ ).
This is just an ordinary differential equation from the point of view of
t
, and
is of the same type of initial value problem that we studied earlier, and the
solution is
˜
G
n
(k, t, y, τ ) = Θ(t τ)e
ik·y
sin |k|c(t τ )
|k|c
.
To recover the Green’s function itself, we have to compute the inverse Fourier
transform, and find
G
n
(x, t; y, τ ) =
1
(2π)
n
Z
R
n
e
ik·x
Θ(t τ )e
ik·y
sin |k|c(t τ )
|k|c
d
n
k.
Unlike the case of the heat equation, the form of the answer we get here does
depend on the number of spatial dimensions
n
. For definiteness, we look at the
case where
n
= 3, since our world (probably) has three dimensions. Then our
Green’s function is
G(x, t; y, τ ) =
Θ(t τ )
(2π)
3
c
Z
R
3
e
ik·(xy)
sin |k|c(t τ )
|k|
d
3
k.
We use spherical polar coordinates with the
z
-axis in
k
-space aligned along the
direction of x y. Hence k · (x y) = kr cos θ, where r = |x y| and k = |k|.
Note that nothing in our integral depends on
ϕ
, so we can pull out a factor
of 2π, and get
G(x, t; y, τ ) =
Θ(t τ )
(2π)
2
c
Z
0
Z
π
0
e
ikr cos θ
sin kc(t τ)
k
k
2
sin θ dθ dk
The next integral to do is the
θ
integral, which is straightforward since it is an
exact differential. Setting α = c(t τ), we get
=
Θ(t τ )
(2π)
2
c
Z
0
e
ikr
e
ikr
ikr
sin kc(t τ)
k
k
2
dk
=
Θ(t τ )
(2π)
2
icr
Z
0
e
ikr
sin kα dk
Z
0
e
ikr
sin kα dk
=
Θ(t τ )
2πicr
1
2π
Z
−∞
e
ikr
sin kα dk
=
Θ(t τ )
2πicr
F
1
[sin kα],
Now recall F[δ(x α)] = e
ikα
. So
F
1
[sin kα] = F
1
e
ikα
e
ikα
2i
=
1
2i
[δ(x + α) δ(x α)].
Hence our Green’s function is
G(x, t; y, τ ) =
Θ(t τ )
4πc|x y|
h
δ
|x y|+ c(t τ)
δ
|x y| c(t τ)
i
.
Now we look at our delta functions. The step function is non-zero only if
t > τ
.
Hence
|x y|
+
c
(
t τ
) is always positive. So
δ
(
|x y|
+
c
(
t τ
)) does not
contribute. On the other hand,
δ
(
|x y|c
(
t τ
)) is non-zero only if
t > τ
. So
Θ(
t τ
) is always positive in this region. So we can write our Green’s function
as
G(x, t; y, τ ) =
1
4πc
1
|x y|
δ(|x y| c(t τ )).
As always, given our Green’s function, the general solution to the forced equation
2
φ
t
2
c
2
2
φ = F (x, t)
is
φ(x, t) =
Z
0
Z
R
3
F (y, τ )
4πc|x y|
δ(|x y| c(t τ )) d
3
y dτ.
We can use the delta function to do one of the integrals. It is up to us which
integral we do, but we pick the time integral to do. Then we get
φ(x, t) =
1
4πc
2
Z
R
3
F (y, t
ret
)
|x y|
d
3
y,
where
t
ret
= t
|x y|
c
.
This shows that the effect of the forcing term at some point
y R
3
affects
the solution
φ
at some other point
x
not instantaneously, but only after time
|x y|/c
has elapsed. This is just as we saw for characteristics. This, again,
tells us that information travels at speed c.
Also, we see the effect of the forcing term gets weaker and weaker as we
move further away from the source. This dispersion depends on the number of
dimensions of the space. As we spread out in a three-dimensional space, the
“energy” from the forcing term has to be distributed over a larger sphere, and
the effect diminishes. On the contrary, in one-dimensional space, there is no
spreading out to do, and we don’t have this reduction. In fact, in one dimensions,
we get
φ(x, t) =
Z
t
0
Z
R
F (y, τ )
Θ(c(t τ ) |x y|)
2c
dy dτ.
We see there is now no suppression factor at the bottom, as expected.
7.5 Poisson’s equation
Let φ : R
3
R satisfy the Poisson’s equation
2
φ = F,
where F (x) is a forcing term.
The fundamental solution to this equation is defined to be G
3
(x, y), where
2
G
3
(x, y) = δ
(3)
(x y).
By rotational symmetry, G
3
(x, y) = G
3
(|x y|). Integrating over a ball
B
r
= {|x y| r, x R
3
},
we have
1 =
Z
B
r
2
G
3
dV
=
Z
B
r
n · G
3
dS
=
Z
S
2
dG
3
dr
r
2
sin θ dθ dφ
= 4πr
2
dG
3
dr
.
So we know
dG
3
dr
=
1
4πr
2
,
and hence
G
3
(x, y) =
1
4π|x y|
+ c.
We often set c = 0 such that
lim
|x|→∞
G
3
= 0.
Green’s identities
To make use of this fundamental solution in solving Poisson’s equation, we first
obtain some useful identities.
Suppose
φ, ψ
:
R
3
R
are both smooth everywhere in some region
R
3
with boundary Ω. Then
Z
φn · ψ dS =
Z
· (φψ) dV =
Z
φ
2
ψ + (φ) ·(ψ) dV.
So we get
Proposition (Green’s first identity).
Z
φn · ψ dS =
Z
φ
2
ψ + (φ) ·(ψ) dV.
Of course, this is just an easy consequence of the divergence theorem, but
when Green first came up with this, divergence theorem hasn’t existed yet.
Similarly, we obtain
Z
ψn · φ dS =
Z
ψ
2
φ + (φ) ·(ψ) dV.
Subtracting these two equations gives
Proposition (Green’s second identity).
Z
φ
2
ψ ψ
2
φ dV =
Z
φn · ψ ψn · φ dS.
Why is this useful? On the left, we have things like
2
ψ
and
2
φ
. These
are things we are given by Poisson’s or Laplace’s equation. On the right, we
have things on the boundary, and these are often the boundary conditions we
are given. So this can be rather helpful when we have to solve these equations.
Using the Green’s function
We wish to apply this result to the case
ψ
=
G
3
(
|x y|
). However, recall that
when deriving Green’s identity, we assumed
φ
and
ψ
are smooth everywhere in
our domain. However, our Green’s function is singular at
x
=
y
, but we want to
integrate over this region as well. So we need to do this carefully. Because of
the singularity, we want to take
Ω = B
r
B
ε
= {x R
3
: ε |x y| R}.
In other words, we remove a small region of radius
ε
centered on
y
from the
domain.
In this choice of Ω, it is completely safe to use Green’s identity, since our
Green’s function is certainly regular everywhere in this Ω. First note that since
2
G
3
= 0 everywhere except at x = y, we get
Z
φ
2
G
3
G
3
2
φ dV =
Z
G
3
2
φ dV
Then Green’s second identity gives
Z
G
3
2
φ dV =
Z
S
2
r
φ(n · G
3
) G
3
(n · φ) dS
+
Z
S
2
ε
φ(n · G
3
) G
3
(n · φ) dS
Note that on the inner boundary, we have n =
ˆ
r. Also, at S
2
ε
, we have
G
3
|
S
2
ε
=
1
4πε
,
dG
3
dr
S
2
ε
=
1
4πε
2
.
So the inner boundary terms are
Z
S
2
ε
φ(n · G
3
) G
3
(n · φ)
ε
2
sin θ dθ dφ
=
ε
2
4πε
2
Z
S
2
ε
φ sin θ dθ dφ +
ε
2
4πε
Z
S
2
ε
(n · φ) sin θ dθ dφ
Now the final integral is bounded by the assumption that
φ
is everywhere smooth.
So as we take the limit
ε
0, the final term vanishes. In the first term, the
ε
’s
cancel. So we are left with
=
1
4π
Z
S
2
ε
φ dΩ
=
¯
φ
φ(y)
where
¯
φ is the average value of φ on the sphere.
Now suppose
2
φ = F . Then this gives
Proposition (Green’s third identity).
φ(y) =
Z
φ(n · G
3
) G
3
(n · φ) dS
Z
G
3
(x, y)F (x) d
3
x.
This expresses
φ
at any point
y
in in terms of the fundamental solution
G
3
, the forcing term
F
and boundary data. In particular, if the boundary values
of φ and n ·φ vanish as we take r , then we have
φ(y) =
Z
R
3
G
3
(x, y)F (x) d
3
x.
So the fundamental solution is the Green’s function for Poisson’s equation on
R
3
.
However, there is a puzzle. Suppose
F
= 0. So
2
φ
= 0. Then Green’s
identity says
φ(y) =
Z
S
2
r
φ
dG
3
dr
G
3
dφ
dr
dS.
But we know there is a unique solution to Laplace’s equation on every bounded
domain once we specify the boundary value
φ|
, or a unique-up-to-constant
solution if we specify the boundary value of n ·φ|
.
However, to get
φ
(
y
) using Green’s identity, we need to know
φ
and and
n · φ on the boundary. This is too much.
Green’s third identity is a valid relation obeyed by solutions to Poisson’s
equation, but it is not constructive. We cannot specify
φ
and
n ·φ
freely. What
we would like is a formula of φ given, say, just the value of φ on the boundary.
Dirichlet Green’s function
To overcome this (in the Dirichlet case), we seek to modify G
3
via
G
3
G = G
3
+ H(x, y),
where
2
H
= 0 everywhere in Ω,
H
is regular throughout Ω, and
G|
= 0. In
other words, we find some
H
that does not affect the relations on
G
when acted
on by
2
, but now our
G
will have boundary value 0. We will find this
H
later,
but given such an
H
, we replace
G
3
with
G H
in Green’s third identity, and
see that all the H terms fall out, i.e. G also satisfies Green’s third identity. So
φ(y) =
Z
[φn · G Gn · φ] dS
Z
F G dV
=
Z
φn · G dS
Z
F G dV.
So as long as we find
H
, we can express the value of
φ
(
y
) in terms of the values
of
φ
on the boundary. Similarly, if we are given a Neumann condition, i.e. the
value of
n · φ
on the boundary, we have to find an
H
that kills off
n · G
on
the boundary, and get a similar result.
In general, finding a harmonic
2
H = 0 with
H|
=
1
4π|x x
0
|
is a difficult problem. However, the method of images allows us to solve this in
some special cases with lots of symmetry.
Example. Suppose
Ω = {(x, y, z) R
3
: z 0}.
We wish to find a solution to
2
=
F
in with
φ
0 rapidly as
|x|
with boundary condition φ(x, y, 0) = g(x, y).
The fundamental solution
G
3
(x, x
0
) =
1
4π
1
|x x
0
|
obeys all the conditions we need except
G
3
|
z=0
=
1
4π
1
[(x x
0
)
2
+ (y y
0
)
2
+ z
2
0
]
1/2
6= 0.
However, let
x
R
0
be the point (
x
0
, y
0
, z
0
) . This is the reflection of
x
0
in the
boundary plane
z
= 0. Since the point
x
R
0
is outside our domain,
G
3
(
x, x
R
0
)
obeys
2
G
3
(x, x
R
0
) = 0
for all x Ω, and also
G
3
(x, x
R
0
)|
z=0
= G
3
(x, x
0
)|
z=0
.
Hence we take
G(x, x
0
) = G
3
(x, x
0
) G
3
(x, x
R
0
).
The outward pointing normal to at z = 0 is n =
ˆ
z. Hence we have
n · G|
z=0
=
1
4π
(z z
0
)
|x x
0
|
3
(z + z
0
)
|x x
R
0
|
3
z=0
=
1
2π
z
0
[(x x
0
)
2
+ (y y
0
)
2
+ z
2
0
]
3/2
.
Therefore our solution is
φ(x
0
) =
1
4π
Z
1
|x x
0
|
1
|x x
R
0
|
F (x) d
3
x
+
z
0
2π
Z
R
2
g(x, y)
[(x x
0
)
2
+ (y y
0
)
2
+ z
2
0
]
3/2
dx dy.
What have we actually done here? The Green’s function
G
3
(
x, x
0
) in some sense
represents a “charge” at
x
0
. We can imagine that the term
G
3
(
x, x
R
0
) represents
the contribution to our solution from a point charge of opposite sign located at
x
R
0
. Then by symmetry, G is zero at z = 0.
z = 0
x
0
x
R
0
Our solution for
φ
(
x
0
) is valid if we extend it to
x
0
R
3
where
φ
(
x
0
) is solved
by choosing two mirror charge distributions. But this is irrelevant for our present
purposes. We are only concerned with finding a solution in the region
z
0.
This is just an artificial device we have in order to solve the problem.
Example.
Suppose a chimney produces smoke such that the density
φ
(
x, t
) of
smoke obeys
t
φ D
2
φ = F (x, t).
The left side is just the heat equation, modelling the diffusion of smoke, while
the right forcing term describes the production of smoke by the chimney.
If this were a problem for x R
3
, then the solution is
φ(x, t) =
Z
t
0
Z
R
3
F (y, τ )S
3
(x y, t τ) d
3
y dτ,
where
S
3
(x y, t τ) =
1
[4πD(t τ)]
3/2
exp
|x y|
2
4D(t τ)
.
This is true only if the smoke can diffuse in all of
R
3
. However, this is not true
for our current circumstances, since smoke does not diffuse into the ground.
To account for this, we should find a Green’s function that obeys
n · G|
z=0
= 0.
This says that there is no smoke diffusing in to the ground.
This is achieved by picking
G(x, t; y, τ ) = Θ(t τ)[S
3
(x y, t τ) + S
3
(x y
R
, t τ )].
We can directly check that this obeys
t
D
2
2
G = δ(t τ)δ
3
(x y)
when x Ω, and also
n · G|z
0
= 0.
Hence the smoke density is given by
φ(x, t) =
Z
t
0
Z
F (y, τ )[S
3
(x y, t τ) + S
3
(x y
R
, t τ )] d
3
y.
We can think of the second term as the contribution from a “mirror chimney”.
Without a mirror chimney, we will have smoke flowing into the ground. With a
mirror chimney, we will have equal amounts of mirror smoke flowing up from
the ground, so there is no net flow. Of course, there are no mirror chimneys in
reality. These are just artifacts we use to find the solution we want.
Example.
Suppose we want to solve the wave equation in the region (
x, t
) such
that x > 0 with boundary conditions
φ(x, 0) = b(x),
t
(x, 0) = 0,
x
φ(0, t) = 0.
On R
1,1
d’Alembert’s solution gives
φ(x, t) =
1
2
[b(x ct) + b(x + ct)]
This is not what we want, since eventually we will have a wave moving past the
x = 0 line.
x = 0
To compensate for this, we introduce a mirror wave moving in the opposite
direction, such that when as they pass through each other at
x
= 0, there is no
net flow across the boundary.
x = 0
More precisely, we include a mirror initial condition
φ
(
x,
0) =
b
(
x
) +
b
(
x
),
where we set
b
(
x
) = 0 when
x <
0. In the region
x >
0 we are interested in, only
the
b
(
x
) term will contribute. In the
x <
0 region, only
x >
0 will contribute.
Then the general solution is
φ(x, t) =
1
2
[b(x + ct) + b(x + c) + b(x ct) + b(x + ct)].