Lecture 3 - The Heat, Wave, and Cauchy-Riemann Equations

Lucas Culler

1 The Heat Equation

Suppose we have a metal ring, and we heat it up in some irregular manner, so that certain parts of it are
hotter than others. Assume the ring is placed in some sort of insulating material, so that no heat is lost to
the environment. How does the temperature at different points in the ring change over time? To answer this
question, suppose that the temperature at time t and at an angle θ relative to a fixed point on the ring is
given by a function u(t, θ). Then the time evolution of u can be modelled by a partial differential equation,
called the “heat equation”:
∂u ∂2u
=κ 2
∂t ∂θ
where κ is a constant depending on the size and physical properties of the ring. As it turns out, this equation
is what motivated Fourier to invent Fourier series, and indeed they are still the best way to solve it. His
method goes like this. First, expand u as a Fourier series with time-dependent coefficients:
X
u(θ, t) = An (t)einθ
n∈Z

Note that the coefficients An can be complex, even though u(θ, t) is assumed to be real. Next, substitute
this expression into the heat equation:
X dAn X
einθ = An κ(in)2 einθ
dt
n∈Z n∈Z

Since Fourier series are unique, we can set these two expressions equal term by term, obtaining (infinitely
many!) ordinary differential equations for the coefficients An :

dAn
= −κn2 An
dt
Solving these equations gives us:
2
An (t) = cn e−κn t

for some constants cn . Substituting back in, we get a formula for u:

X 2
u(θ, t) = cn e−κn t+inθ
n∈Z

So we are left to determine the constants cn . How can we do this? Well, setting t = 0, we get
X
u(θ, 0) = cn einθ
n∈Z

So if we know the initial temperature distribution then we can find the constants cn by taking its Fourier
transform: Z 2π
1
cn = u(θ, 0)e−inθ dθ
2π 0

1
Thus Fourier series provide us with a thorough, systematic way to solve the heat equation with periodic
boundary conditions.
What does this solution tell us physically? Taking the limit as t → ∞ shows that the temperature of the
ring approachs a constant, namely c0 . Moreover, looking at the constants in the exponentials we see that
the higher Fourier modes die off much quicker than the lower modes. So rapid oscillations get smoothed out
rather quickly, whereas more modest oscillations will take a longer time to reach equilibrium.
One peculiar feature of the heat equation is that it cannot be solved infinitely far back in time. The
reason for this is that the Fourier series describing the solution ceases to be convergent when the coefficients
grow too large. However, our argument shows that a solution does exist for all positive times, and this
solution is uniquely determined by the initial conditions.

2 The Wave Equation

Now suppose we take our metal ring and strike it with a hammer. The ring will then vibrate, meaning that
it will be displaced from its original shape, and the displacement will change with time, oscillating back and
forth, presumably at a fairly high frequency. The displacement φ(θ, t) of the ring can be modelled (assuming
dissipative effects are negligible) by the following partial differential equation, called the “wave equation”:
∂2φ ∂2φ
2
= c2 2
∂t ∂θ
where c is some constant depending on things like the density of the ring and its radius. Let’s again use
Fourier series to solve this equation. Since the displacement is periodic, we can write it as follows:
X
φ(θ, t) = An (t)einθ
n∈Z

where An are complex valued functions. We can then plug this into the equation above, obtaining:
X d 2 An X
2
einθ = An (t)c2 (in)2 einθ
dt
n∈Z n∈Z

Setting coefficients equal to one another we obtain:

d2 An
= −c2 n2 An
dt2
In this case, the ordinary differential equation has two independent solutions:

An (t) = an einct + bn e−inct

so that the full solution is given by:

X X
φ(θ, t) = an ein(θ+ct) + bn ein(θ−ct)
n∈Z n∈Z

So we see that φ can be written as a sum of two arbitrary functions

φ(θ, t) = f (θ + ct) + g(θ − ct)

which correspond physically to waves of displacement propogating in the clockwise and counterclockwise
directions.
Even though we can eliminate Fourier series from our answer at the end, the Fourier decomposition of
the wave is still interesting from a physical point of view. In particular, note that the function

φ(θ, t) = ein(θ+ct)

2
returns to its original configuration whenever ct is a multiple of 2π
n . So, the sound waves emitted by the ring
will have frequencies
nc
νn =
2π
Thus there will be a lowest or fundamental frequency, with overtones occuring at integer multiples of that
frequency. Of course, we should really have allowed the ring to have a circumference other than 2π, in which
case the frequencies would be
nc
νn =
L
This shows that if we make the ring wider, then its fundamental frequency decreases. In other words, the
note we hear when we strike it has a lower pitch. Of course, this phenomenon is familiar from our experiences
with musical instruments, and so on.

3 The Cauchy-Riemann Equations

Now that we have seen how to use Fourier series to solve partial differential equations of physical importance,
let’s try to apply the same method to the case we really care about, the Cauchy-Riemann equation. Recall
that this equation is:
∂f ∂f
= −i
∂x ∂y
First let’s try to find all solutions that are periodic in the x variable. We proceed exactly as before. First
we write f as a Fourier series in x: X
f (x, y) = An (y)einx
n∈Z

Then we use the Cauchy-Riemann equations to get ordinary differential equations satisfied by the functions
An (y):
dAn
inAn = −i
dy
Solving these, we get:
An = cn e−ny
and therefore f is given by X X
f (x, y) = cn e−ny+inx = cn ein(x+iy)
n∈Z n∈Z

This can be expressed in a form that makes it evidently holomorphic:

X
f (z) = cn einz
n∈Z

So, periodic holomorphic functions have Fourier series in much the same way as periodic real-valued functions
have Fourier series.
Note that the Fourier series of a periodic holomorphic function will only converge when y is constrained to
lie in some interval, because the positive Fourier coefficients increase exponentially with y and the negative
Fourier coefficients decrease exponentially with y. We will exploit this momentarily when we prove the
Riemann extension theorem and the Liouville theorem.
Theorem. Let f be a holomorphic function defined on an annulus A = {z ∈ C : r1 < |z| < r2 }. Then f is
given on A by a Laurent series (that is, a power series with both positive and negative powers):
X
f (z) = cn z n
n∈Z

Moreover, the coefficients cn are uniquely determined by the function f .

3
Proof. The idea is to use “logarithmic coordinates”. More concretely, consider the infinite horizontal strip

S = (−∞, ∞) × (log r1 , log r2 )

There is a conformal map g : S → A given by z 7→ eiz . Given any holomorphic function f on A, we therefore
get a periodic holomorphic function f ◦ g on S. By the computation above, f ◦ g can be written as follows:
X X
f (g(z)) = cn einz = cn g(z)n
n∈Z n∈Z

for some constants cn . Setting w = g(z), we get:

X
f (w) = cn wn
n∈Z

for all w ∈ A, as claimed.

We often want to know about holomorphic functions on a disk rather than an annulus. The following
result says that in this case the laurent series expansion has only nonnegative powers of z.
Theorem. (Riemann Extension Theorem) Let f be a holomorphic function defined on a punctured disk
D∗ = {z ∈ C : 0 < |z| < r}. Suppose moreover that f is bounded by some constant, i.e. |f (z)| < M for all
z ∈ D∗ . Then f is given by a power series
X
f (z) = cn z n
n≥0

In particular, we can use the power series expansion to extend f to a holomorphic function defined on the
entire disk D, and this extension is unique.

Proof. The preimage of D∗ under eiz is an upper half-space H = {z = x + iy : y > − log r} . Using the
standard formula for Fourier coefficients, we get:
Z 2π
1
cn e−ny = f (x + iy)e−inx dx
2π 0
So if f is bounded by a constant M , then we get:

|cn e−ny | ≤ M

for all y > log r. Hence if n is negative, cn must be zero, which shows that the Laurent expansion of f only
has nonnegative terms.
Corollary 1. Any holomorphic function f defined on an open disk D has a power series expansion valid on
that disk: X
f (z) = cn z n
n≥0

Corollary 2. (Liouville’s Theorem) Any bounded, holomorphic function defined on the entire complex plane
is constant.

Proof. Proceeding as in the proof of the Riemann extension theorem, we see that for every n ∈ Z we must
have
|cn e−ny | ≤ M
for some constant M . Since the expression on the left gets arbitrarily large when y → ±∞, we must have
cn = 0 for all nonzero n. Therefore, f is constant.