Math 180 Spring 2018 Notes

Notes on Partial Differential Equations
Preliminary Lecture Notes
Adolfo J. Rumbos
c Draft date April 10, 2018
2
Contents
1 Preface 5
2 Modeling with PDEs 7

2.1 Modeling Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 The Continuity Equation . . . . . . . . . . . . . . . . . . 9
2.1.2 Conservation of Momentum . . . . . . . . . . . . . . . . . 17
2.1.3 Conservation of Energy . . . . . . . . . . . . . . . . . . . 21
2.1.4 Euler Equations . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Modeling Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Variational Problems . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1 Minimal Surfaces . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.2 Linearized Minimal Surface Equation . . . . . . . . . . . 33
2.3.3 Vibrating String . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Modeling Small Amplitude Vibrations . . . . . . . . . . . . . . . 39
3 Classification of PDEs 43
3.1 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 Second Order PDEs . . . . . . . . . . . . . . . . . . . . . . . . . 46
4 Solving PDEs 51
4.1 Using Characteristic Curves . . . . . . . . . . . . . . . . . . . . . 51
4.1.1 Solving the One–Dimensional Wave Equation . . . . . . . 52
4.1.2 Solving First–Order PDEs . . . . . . . . . . . . . . . . . . 57
4.2 Using Symmetry to Solve PDEs . . . . . . . . . . . . . . . . . . . 64
4.2.1 Radially Symmetric Solutions . . . . . . . . . . . . . . . . 64
4.2.2 Dilation-Invariant Solutions . . . . . . . . . . . . . . . . . 69
4.2.3 Solving the Diffusion Equation . . . . . . . . . . . . . . . 73
5 Solving Linear PDEs 77

5.1 Solving the Vibrating String Equation . . . . . . . . . . . . . . . 77
5.1.1 Separation of Variables . . . . . . . . . . . . . . . . . . . 79
5.1.2 Fourier Series Expansions . . . . . . . . . . . . . . . . . . 85
5.1.3 Differentiability of Fourier Series . . . . . . . . . . . . . . 100
5.1.4 Solution of the Vibrating String Problem . . . . . . . . . 116
3
4 CONTENTS
5.2 Fundamental Solutions . . . . . . . . . . . . . . . . . . . . . . . . 119

5.2.1 Heat Kernel . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2.2 Uniqueness for the Diffusion Equation . . . . . . . . . . . 134
5.3 Dirichlet Problem for the Unit Disc . . . . . . . . . . . . . . . . . 140
5.3.1 Separation of Variables . . . . . . . . . . . . . . . . . . . 141
5.3.2 An Eigenvalue Problem . . . . . . . . . . . . . . . . . . . 143
5.3.3 Expansion in Terms of Eigenfunctions . . . . . . . . . . . 148
5.3.4 The Poisson Kernel . . . . . . . . . . . . . . . . . . . . . 156
5.3.5 The Poisson Integral . . . . . . . . . . . . . . . . . . . . . 161
5.3.6 Existence for the Dirichlet Problem on the Disc . . . . . . 167
5.4 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.4.1 Green’s Integral Representation Formula . . . . . . . . . . 170
5.4.2 Definition of Green’s Function . . . . . . . . . . . . . . . 176
5.4.3 Solving Poisson’s Equation . . . . . . . . . . . . . . . . . 183
A Facts from the Theory of ODEs 191

A.1 Linear, Second Order ODEs with Constant Coefficients . . . . . 191
B Theorems About Integration 193

B.1 Differentiating Under the Integral Sign . . . . . . . . . . . . . . . 193
B.2 The Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . 194
C Kernels 197
C.1 The Dirichlet Kernel . . . . . . . . . . . . . . . . . . . . . . . . . 197
Chapter 1
Preface
This course is an introduction to the theory and applications of partial differ-

ential equations (PDEs). PDEs are expressions involving functions of several
variables and its partial derivatives in which we seek to find one of the func-
tions, or a set of functions, subject to some initial conditions (if time is involved
as one of the variables) or boundary conditions. They arise naturally when
modeling physical or biological systems in which assumptions of continuity and
differentiability are made about the quantities in question. In these notes we
will discuss several modeling situations that give rise to PDEs.
In problems involving PDEs we are mainly interested in the question of ex-
istence of solutions. In some cases, answering this questions amounts to coming
up with formulas for the solutions. In these notes we will discuss a few tech-
niques for constructing solutions (e.g., separation of variables, series expansions
and Green’s function methods) for the special case of linear equations.
Linear partial differential equations are very important because they come
up in many applications in the natural sciences. There are three major classes of
linear PDEs: hyperbolic, parabolic and elliptic equations. Archetypal instances
of these classes of PDEs are the classical equations of mathematical physics:
the wave equation, the heat or diffusion equations, and Laplace’s or Poisson’s
equations, respectively. In simple instances of these equations in one and two–
dimensional space, we will show how to construct solutions subject to some
initial conditions and some boundary conditions. These constructions will be
based on the method of Fourier series expansions. We will also explore other
methods for constructing solutions involving Green’s functions and transform
methods. All these methods rely on the linear structure of the equations.
In most cases, however, explicit constructions of solutions are not possible.
In these cases, the only recourse we have is analytical proofs of existence, or
nonexistence, and qualitative analysis to deduce properties of solutions. Once
an existence theorem is obtained for a particular PDE problem, the next step
in the analysis might involve approximation techniques to get information on
the behavior and property of solutions. In these notes, we will discuss a few of
those computational techniques.
5
6 CHAPTER 1. PREFACE
Chapter 2
How Do PDEs Arise?
In general, a partial differential equation for a function, u, of several variables,

u(x1 , x2 , . . . , xn ), is an expression of the form
F (x, u, ux1 , . . . , uxn , ux1 x2 , . . . , uxn xn , . . .) = 0, (2.1)
where x = (x1 , . . . , xn ) and ux1 , . . . , uxn , ux1 x2 , . . . , uxn xn , . . . denote partial

derivatives of u, for some function, F , of several variables. For example, in
the simplest case in which u is a function of time, t ∈ R, and a single space
variable x ∈ R, an instance of (2.1) is provided by
ut − kuxx = 0, (2.2)
for some constant k.

While we are usually interested in knowing when equations like (2.1) and
(2.2) have solutions subject to some initial and/or boundary conditions, in this
chapter we will focus on the questions of how those equations arise in practice.
For instance, the equation in (2.2) describes one–dimensional heat flow (u(x, t) in
this case denotes the temperature at time t and location x), or one–dimensional
diffusion (u(x, t) denotes the concentration of a substance at time t and location
x). We begin by deriving a system of PDEs that describe the motion of fluids.
2.1 Modeling Fluid Flow

In this section we illustrate the use of a very important modeling principle,
which we shall refer to as a conservation principle. This is a rather general
principle that can be applied in situations in which the evolution in time of the
quantity of a certain entity within a certain system is studied. For instance,
suppose the quantity of a certain substance confined within a system is given
by a continuous function of time, t, and is denoted by Q(t) (the assumption
of continuity is one that needs to be justified by the situation at hand). A
conservation principle states that the rate at which a the quantity Q(t) changes
7
8 CHAPTER 2. MODELING WITH PDES
has to be accounted by how much of the substance goes into the system and
how much of it goes out of the system. For the case in which Q is also assumed
to be differentiable (again, this is a mathematical assumption that would need
some justification), the conservation principle can be succinctly stated as
dQ
= Rate of Q in − Rate of Q out. (2.3)
dt
In the cases to be considered in this section, the conservation principle in (2.3)
might lead to a differential equation, or a system of differential equations, and
so the theory of differential equations will be used to help in the analysis of the
model.
In the derivation of the equations governing fluid motion, we will have the
opportunity to apply the conservation principle in (2.3) several times.
Suppose we are following the motion of a fluid in some region R in three–
dimensional space; see Figure 2.1.1. We assume that the fluid is a continuum
(x(t), y(t), z(t))
(x, y, z)
B
Figure 2.1.1: Region R
with density function ρ(x, y, x, t), in units of mass per unit volume, so that the
mass of a fluid element of volume dV = dxdydz around a point (x, y, z) at time
t is, approximately,
ρ(x, y, x, t)dV,
where dV denotes the volume of the fluid element. It then follows that the mass
of fluid contained in a subregion B of R (see Figure 2.1.1) at time t is given by
ZZZ
M (B, t) = ρ(x, y, x, t) dV. (2.4)
B
We assume throughout this discussion that ρ is a continuous function.

We also assume that each fluid element located at (x, y, z) at time t moves
according to a velocity vector ~u = (u1 , u2 , u3 ), where u1 , u2 and u3 are dif-
ferentiable functions of (x, y, z, t). Thus, the path that a fluid element located
2.1. MODELING FLUID FLOW 9
at (x, y, z) at time t = 0 will follow is determined by the following system of

ordinary differential equations

dx

 = u1 (x(t), y(t), z(t), t);
dt







dy

= u2 (x(t), y(t), z(t), t); (2.5)
 dt





 dz = u3 (x(t), y(t), z(t), t),



dt
subject to the initial conditions

 x(0) = x;

y(0) = y; (2.6)

z(0) = z.

If we assume that the components of the velocity field ~u are differentiable with
continuous derivatives throughout the region R and for all times t (i.e., ~u is a
C 1 vector field), then a solution to the system of ordinary differential equations
in (2.5) subject to the initial conditions in (2.6) is guaranteed to exist over
some maximal interval of time containing 0. The solution (x(t), y(t), y(t)) of
the system in (2.5) subject to the initial conditions in (2.6) defines a path in
space,
t 7→ (x(t), y(t), y(t)),
for t in the maximal interval of existence, which describes the motion of a fluid
element located at (x, y, z) at time t = 0. The path traced by the fluid element
as it moves in time is called a pathline; Figure 2.1.1 shows what a pathline
through (x, y, z) might look like. If we knew the velocity field at any point
in space and at any time, we could compute the pathline through (x, y, z) by
integrating the equations in (2.5) and imposing the initial conditions in (2.6):
Z t
x(t) = x + u1 (x(τ ), y(τ ), z(τ ), τ ) dτ ;
0
Z t
y(t) = y+ u2 (x(τ ), y(τ ), z(τ ), τ ) dτ ; (2.7)
0
Z t
z(t) = z+ u3 (x(τ ), y(τ ), z(τ ), τ ) dτ.
0
However, the velocity field is usually not known, and we need to do more mod-
eling to find equations involving u1 , u2 and u3 that we hope we can solve.
2.1.1 The Continuity Equation

Consider a subregion, B, of R, with smooth boundary ∂B, as that pictured in
Figure 2.1.1. The mass of the fluid contained at time t in that region, MB (t),
is given by equation (2.4),

ZZZ
MB (t) = ρ(x, y, x, t) dV. (2.8)
B
By the principle of conservation of mass, the rate of change in the mass of fluid
contained in B has to be accounted for by how much fluid is entering the region
and how much is leaving per unit of time:
dMB
= Rate of fluid into B − Rate of fluid out of B. (2.9)
dt
The equation in (2.9) is an instance of the conservation principle in (2.3).
If we assume that ρ is a C 1 function in R, we can compute the left–hand
side of the equation by differentiating under the integral in (2.8):
ZZZ
dMB ∂ρ
= (x, y, x, t) dV. (2.10)
dt B ∂t
Next, we compute the right–hand side of the expression in (2.9). Let ~n denote
the unit vector normal to the boundary, ∂B, of the region B pointing outward.
The outward unit normal, ~n(x, y, z), to the boundary of B is guaranteed to exist
at every point (x, y, z) ∈ ∂B if we assume that ∂B is a smooth surface. Then,
the rate of fluid passing through an element of area, dA, on the surface ∂B can
be expressed, approximately, as
ρ ~u · ~n dA, (2.11)
where ~u · ~n denotes the dot product of ~u and ~n. Note that the expression in
(2.11) is in units of mass per unit of time. Integrating the expression in (2.11)
over the boundary of B yields the net flux of mass across the surface ∂B,
ZZ
ρ ~u · ~n dA. (2.12)
∂B
Since the outward unit normal, ~n, points away from the region B, the expression
in (2.12) measures the flux of fluid away from the region B, if it is positive; if
the expression in (2.12) is negative, it measures the net amount of fluid per unit
time that enters B. We can therefore write the conservation principle in (2.9)
as ZZ
dMB
=− ρ ~u · ~n dA. (2.13)
dt ∂B
To understand the reason for the minus sign on the right–hand side of the
expression in (2.13), observe that a net increase in the amount of fluid in the
region B, which yields a positive sign for the derivative in the left–hand side of
(2.13), corresponds to a net amount of fluid flowing into the region B across the
boundary ∂B.
Since we are assuming that the boundary of B is smooth, we can apply the
Divergence Theorem to rewrite the integral in the right–hand side of (2.13) as
follows: ZZ ZZZ
ρ ~u · ~n dA = ∇ · (ρ~u) dV, (2.14)
∂B B
where ∇ · (ρ~u) denotes the divergence of the vector field ρ~u; that is,
∂ ∂ ∂
∇ · (ρ~u) = (ρu1 ) + (ρu2 ) + (ρu3 ). (2.15)
∂x ∂y ∂z
In view of (2.10) and (2.14), we see that we can rewrite the conservation
equation in (2.13) as
ZZZ ZZZ
∂ρ
dV = − ∇ · (ρ~u) dV,
B ∂t B
or ZZZ
∂ρ
+ ∇ · (ρ~u) dV = 0. (2.16)
B ∂t
If we assume that the vector field ~u and the scalar field ρ are C 1 functions over
R and for all times t, then the fact that (2.16) holds true for any subregion B of
R with smooth boundary implies that integrand on the left–hand side of (2.16)
must be 0 over R and for all t; that is,
∂ρ
+ ∇ · (ρ~u) = 0, in R and for all t. (2.17)
∂t
The equation in (2.17) is an example of a partial differential equation (PDE)
involving the functions ρ, u1 , u2 and u3 ; in fact, using the definition of divergence
(see (2.15)), the PDE in (2.17) can be rewritten as
∂ρ ∂ ∂ ∂
+ (ρu1 ) + (ρu2 ) + (ρu3 ) = 0. (2.18)
∂t ∂x ∂y ∂z
The PDE in (2.17) is called the continuity equation and it expresses the
conservation principle for a quantity of density ρ that flows according to a veloc-
ity field ~u in some region in space. For one–dimensional flow with linear density
ρ(x, t) and scalar velocity field u(x, t), for x ∈ R and t ∈ R, the continuity
equation reads
∂ρ ∂
+ (ρu) = 0; (2.19)
∂t ∂x
see (2.18). The equation in (2.19) is an example of a first order PDE because
the first derivatives of the functions ρ and u are involved. As it stands, the PDE
in (2.19) involves two unknown functions, the density, ρ, and the velocity, u.
Thus, we will need one more relation or equations in order for us to even begin
to solve the problem posed by the modeling that led to the PDE in (2.19). An
interesting example is provided by the following application to modeling traffic
flow.
Example 2.1.1 (Modeling Traffic Flow). Consider the unidirectional flow of

traffic in a one–lane, straight road depicted in Figure 2.1.2. In this idealized
road, vehicles are modeled by moving points. The location, x, of a point–vehicle
is measured from some reference point along an axis parallel to the road. We
u1r - u2 r -
Figure 2.1.2: One–lane unidirectional flow
postulate a traffic density, ρ(x, t), measured in units of number of cars per unit
length of road at location x and time t. We interpret ρ(x, t) as follows: Consider
a section of the road from x to x + ∆x at time t. Let ∆N ([x, x + ∆x], t) denote
the number of cars in the section [x, x + ∆x] at time t. We define ρ(x, t) by the
expression
∆N ([x, x + ∆x], t)
ρ(x, t) = lim , (2.20)
∆x→0 ∆x
provided that the limit on the right–hand side of (2.20) exists. It follows from
(2.20) that, if a continuous traffic density, ρ(x, t), is known for all x and t, then
the number of cars in a section of the road from x = a to x = b, where a < b,
at time t is given by
Z b
∆N ([a, b], t) = ρ(x, t) dx.
a
We assume that at each point x along the road and at each time t the velocity
of vehicle at that location and time is dictated by a function u(x, t), which we
also assume to be a C 1 function. It follows from these assumptions and the
derivations in this section that the one–dimensional equation of continuity in
(2.19) applies to this situation.
Ideally, we would like to find a solution, ρ, to (2.19) subject to some initial
condition
ρ(x, 0) = ρo (x), (2.21)
for some initial traffic density profile, ρo , along the road. In order to solve this
problem, we postulate that u is a function of traffic density—the higher the
density, the lower the traffic speed, for example. We may therefore write
u = f (ρ, Λ), (2.22)
where f is a continuous function of ρ and a set of parameters, Λ. Some of the

parameters might be a maximum density, ρmax , dictated by bumper to bumper
traffic, and a maximum speed, vmax ; for instance, vmax is a speed limit. Given
the parameters ρmax and vmax , the simplest model for the relationship between
v and ρ is the constitutive equation

ρ
u = vmax 1 − . (2.23)
ρmax
We therefore arrive at the initial value problem (IVP):

∂ρ ∂ ρ
+ vmax ρ 1− = 0 for x ∈ R, t > 0;


∂t ∂x ρmax

(2.24)


ρ(x, 0) = ρo (x), for x ∈ R,

where we have incorporated the continuity equation in (2.19), the initial con-
dition in (2.21), and the constitutive relation in (2.23), which is an instance of
(2.22).
The partial differential equation model for traffic flow (2.24) presented in this
section, based on the equation of continuity in (2.19) and a constitutive relation
for the traffic velocity, u, and the traffic density ρ (of which (2.23) is just an
example), was first introduced by Lighthill and Whitman in 1955 (see [LW55]);
it was also treated by Richards in 1956, [Ric56]. In a subsequent section in
these notes we will present an analysis of this model based on the method of
characteristics.
We end this section with an alternate derivation of the conservation of mass
equation in (2.17). In this approach we focus on the amount of fluid contained
in a region B as the fluid in this region moves according to flow dictated by
the velocity field ~u. Suppose we begin to observe a portion of fluid in B at
time t = 0. We assume that B is bounded and has smooth boundary ∂B. At
some time t > 0, the portion of fluid in B has moved as a consequence of the
fluid motion. We denote by Bt the portion of the fluid that we are following at
time t (see Figure 2.1.3). To see how Bt comes about, consider a fluid element
located at (x, y, z) at time t = 0. At time t > 0, the fluid element will be located
at (x(t), y(t), z(t)), where the functions x(t), y(t) and z(t) are solutions to the
system of ordinary differential equations in (2.5) subject to the initial conditions
in (2.6). We denote the point (x(t), y(t), z(t)) by ϕt (x, y, z), and note that the
map
(x, y, z) 7→ ϕt (x, y, z), for all (x, y, z) ∈ R,
yields a C 1 map from R to R. Furthermore, ϕt is an invertible map for each t
in the interval of existence for the initial value problem in (2.5) and (2.6). We
shall refer to ϕt as the fluid flow map; it gives the location of a fluid element
initially at (x, y, y) at time t as a result of fluid motion. It then follows that Bt
is the image of B under the flow map ϕt ; that is,
Bt = ϕt (B). (2.25)
The total mass of the fluid in Bt is a function of time that we compute as follows
ZZZ
m(t) = ρ(ϕt (x, y, z), t) dV. (2.26)
Bt
Bt
Figure 2.1.3: Balance of Forces
Note that ZZZ

m(0) = ρ(x, y, y, 0) dV ≡ mo , (2.27)
B
which is the mass of the portion of fluid in the region B. As the flow of the fluid
moves the region B, its shape might change. However, because of conservation
of mass, the mass of fluid contained in Bt must the same as that contained in
the region B at time t = 0; that is,
m(t) = mo , for all t, (2.28)
where mo is the constant given in (2.27). It follows from (2.28) that
dm
= 0, for all t. (2.29)
dt
dm
Before we compute , we first rewrite the integral defining m(t) in (2.26)
dt
by means of the change of variables provided by the flow map ϕt (see (2.25).
We have ZZZ
m(t) = ρ(x, y, z, t)J(x, y, z, t) dxdydz
B
where J(x, y, z, t) the Jacobian of the map ϕt ; that is, J(x, y, z, t) is the deter-
minant of the derivative map of ϕt . We then have that
ZZZ
dm ∂
= [ρJ] dxdydz,
dt B ∂t
or ZZZ
dm ∂J ∂ρ
= ρ + J dxdydz. (2.30)
dt B ∂t ∂t
Making the change of variables provided by the flow map in the integral in (2.30)
we obtain that
ZZZ
dm 1 ∂
= ρ(ϕt (x, y, z), t) J(ϕt (x, y, z), t)
dt Bt J(ϕt (x, y, z), t) ∂t
(2.31)
∂
+ [ρ(ϕt (x, y, z), t)] dV.
∂t
It can be shown that
∂
[J(ϕt (x, y, z), t) = J(ϕt (x, y, z), t) ∇ · ~u(ϕt (x, y, z), t), (2.32)
∂t
see page 8 in [CM93]. Thus, substituting (2.32) into (2.31), we get
ZZZ
dm ∂
= (∇ · ~u(ϕt (x, y, z), t))ρ(ϕt (x, y, z), t) + [ρ(ϕt (x, y, z), t)] dV,
dt Bt ∂t
which we can write as
ZZZ
dm Dρ
= (∇ · ~u)ρ + dV, (2.33)
dt Bt Dt
where we have set
Dρ ∂
= [ρ(ϕt (x, y, z), t)]
Dt ∂t
∂
= [ρ(x(t), y(t), z(t), t)] (2.34)
∂t
∂ρ dx ∂ρ dy ∂ρ dz ∂ρ
= + + + ,
∂x dt ∂y dt ∂z dt ∂t
where we have used the Chain Rule in the last step of the calculations in (2.34)
and assumed that the density ρ is a C 1 field. We therefore have that
∂ ∂ρ ∂ρ ∂ρ ∂ρ
[ρ(x(t), y(t), z(t), t)] = + u1 + u2 + u3 , (2.35)
∂t ∂t ∂x ∂y ∂z
where we have used the fact that (x(t), y(t), z(t)) solves the system of ordinary
differential equations in (2.5). Writing (2.35) in vector notation we obtain
∂ ∂ρ
[ρ(x(t), y(t), z(t), t)] = + ~u · ∇ρ, (2.36)
∂t ∂t

∂ρ ∂ρ ∂ρ
where ∇ρ = , , is the gradient of ρ. The expression in (2.36) is
∂x ∂y ∂z
called the material derivative of the field ρ. It is also referred to as the
Dρ
convective derivative of ρ and is usually denoted by , so that
Dt
Dρ ∂ρ
= + ~u · ∇ρ. (2.37)
Dt ∂t
In general, given a C 1 scalar field, g, defined in a region R, the material

derivative of g is given by
Dg ∂g
= + ~u · ∇g. (2.38)
Dt ∂t
The material derivative of g in (2.38) expresses the rate of change of g along the
pathlines as a result of the fact that the field g might change in time as well as
a result of the motion of the fluid. The material derivative of a C 1 vector field,
→
−
G = (g1 , g2 , g3 ), is
→
−
DG Dg1 Dg2 Dg3
= , , ,
Dt Dt Dt Dt
which can be written as
→
− →
−
DG ∂G →
−
= + (~u · ∇) G . (2.39)
Dt ∂t
Combining (2.29) with (2.33) we get that

ZZZ
Dρ
(∇ · ~u)ρ + dV = 0, for all t. (2.40)
Bt Dt
It follows from (2.40) that
Dρ
+ (∇ · ~u)ρ = 0, in R, for all t, (2.41)
Dt
Dρ
where the material derivative, , of ρ is given in (2.37); that is,
Dt
Dρ ∂ρ
= + ~u · ∇ρ
Dt ∂t
(2.42)
∂ρ
= + ∇ · (ρ~u) − (∇ · ~u)ρ
∂t
substituting the result of the calculations in (2.42) into (2.41) then yields
∂ρ
+ ∇ · (ρ~u) = 0, in R, for all t,
∂t
which is the continuity equation in (2.17). We have also shown that the equation
in (2.41) is an equivalent form of the continuity equation. We shall rewrite it
here as
Dρ
= −(∇ · ~u)ρ, in R, for all t. (2.43)
Dt
2.1.2 Conservation of Momentum for an Ideal Fluid

The total momentum at time t of a the portion of fluid contained in a region
Bt with smooth boundary, ∂Bt , is given by
→
−
ZZZ
Π B (t) = ρ(x(t), y(t), z(t), t)~u(x(t), y(t), z(t), t) dV,
Bt
or
→
−
ZZZ
Π B (t) = ρ(ϕt (x, y, x), t)~u(ϕt (x, y, x), t) dV,
Bt
and which we’ll simply write as
→
−
ZZZ
Π B (t) = ρ~u dV, (2.44)
Bt
(see Figure 2.1.3). The principle of conservation of momentum states that the
rate of change of the total momentum of the fluid in Bt has to be accounted for
by the balance of forces acting on Bt :
→
−
dΠB
= Balance of Forces on Bt ; (2.45)
dt
this is, in fact, Newton’s second law of motion.
There are two types of forces acting on the portion of fluid in Bt that con-
tribute to the balance of forces in the right–hand side of the equation in (2.45).
There are forces of stress due to the fluid surrounding the region Bt , and there
are external, or body forces, such as gravity or electromagnetic forces. We can
then rewrite the conservation of momentum equation in (2.45) as
→
−
dΠB →
− →
−
= S B (t) + F B (t), (2.46)
dt
→
−
where S B (t) denotes the total vector sum of the stress forces acting on Bt , and
→
−
F B (t) the total vector sum of body forces acting on Bt .
We assume that
→
−
ZZZ
F B (t) = f~ dV, (2.47)
Bt
where the vector field f~(x, y, z, t) gives the total forces per unit volume acting
on an element of fluid around the point (x, y, z) at time t.
In this section we shall make a special assumption when modeling the stress
forces acting on the fluid. We assume that the fluid under consideration is an
ideal fluid. This means that at any point, (x, y, y), on a surface in the fluid,
the stress force per unit area exerted across the surface is given by
p(x, y, z, t)~n
where ~n is a unit vector perpendicular to the surface at (x, y, z) and time t, and
p(x, y, z, t) is a scalar field called the pressure. It then follows that
→
−
ZZ
S B (t) = − p~n dA, (2.48)
∂Bt
where ~n is the outward unit normal to ∂Bt .

Substituting the expressions in (2.48) and (2.47) into the conservation of
momentum expression in (2.46) yields
→
− ZZ ZZZ
dΠB
=− p~n dA + f~ dV. (2.49)
dt ∂Bt Bt
Writing the unit vector ~n in Cartesian coordinates, (n1 , n2 , n3 ), we see that the
stress forces term in (2.49) has components
ZZ ZZ ZZ
− pn1 dA, − pn2 dA, and − pn3 dA.
∂Bt ∂Bt ∂Bt
Applying the divergence theorem to each of these components we get

ZZ ZZZ
∂p
− pn1 dA = − dV
∂Bt Bt ∂x
ZZ ZZZ
∂p
− pn2 dA = − dV
∂Bt Bt ∂y
and ZZ ZZZ
∂p
− pn3 dA = − dV.
∂Bt Bt ∂z
→
−
Substituting these expressions into the definition of S B (t) in (2.48) we obtain
ZZZ
~B (t) = −
S ∇p dV, (2.50)
Bt
where
∂p ∂p ∂p
∇p = , , (2.51)
∂x ∂y ∂z
is the gradient of p. Combining (2.50), (2.48) and (2.46), we can rewrite the
conservation of momentum equation in (2.49) as
→
− ZZZ ZZZ
dΠB
=− ∇p dV + f~ dV, (2.52)
dt Bt Bt
where ∇p is as given in (2.51).

Next, we see how to compute the left–hand side of the equation in (2.52),
→
− ZZZ
dΠB d
= ρ~u dV, (2.53)
dt dt Bt
according to the definition of momentum in (2.44).

Observe that, since Bt comes about as the result of the action of the flow
map ϕt on B (see 2.25), we can rewrite the integral on the right–hand side of
(2.53) as
ZZZ ZZZ
ρ~u dV = ρ(ϕt (x, y, z), t)~u(ϕt (x, y, z), t) dV
Bt Bt
ZZZ
= ρ(x, y, z, t)~u(x, y, z, t)J(x, y, z, t) dxdydz
B
where J(x, y, z, t) the Jacobian of the map ϕt ; that is, J(x, y, z, t) is the deter-
minant of the derivative map of ϕt . We then have that
ZZZ ZZZ
d ∂
ρ~u dV = [Jρ~u] dxdydz,
dt Bt B ∂t
or ZZZ ZZZ
d ∂J ∂
ρ~u dV = ρ~u + [ρ~u]J dxdydz. (2.54)
dt Bt B ∂t ∂t
Substituting (2.32) into (2.54) yields
ZZZ ZZZ
d ∂
ρ~u dV = (∇ · ~u)ρ~u + [ρ~u] J dxdydz,
dt Bt B ∂t

ZZZ ZZZ
d ∂
ρ~u dV = (∇ · ~u)ρ~u + [ρ~u] dV. (2.55)
dt Bt Bt ∂t
Using the expression for the material derivative of a vector field in (2.39), we
can rewrite (2.55) as
ZZZ ZZZ
d D
ρ~u dV = (ρ~u) + (∇ · ~u)ρ~u dV. (2.56)
dt Bt Bt Dt
Using the definition of the convective derivative for a vector field in (2.39) we
have that
D D~u Dρ
(ρ~u) = ρ + ~u, (2.57)
Dt Dt Dt
where
Dρ ∂ρ
= + ~u · ∇ρ
Dt ∂t
∂ρ
+ ∇ · (ρ~u) − ρ∇ · ~u;
=
∂t
it then follows from the conservation mass equation in (2.17) that
Dρ
= −ρ∇ · ~u, (2.58)
Dt
which is an alternate form of the conservation of mass principle.

Combining (2.57) and (2.58) then yields
D D~u
(ρ~u) = ρ − (∇ · ~u)ρ~u. (2.59)
Dt Dt
D
Substituting the expression for (ρ~u) in (2.59) into the expression for the
Dt
rate of change of momentum in (2.56) yields
ZZZ ZZZ
d D~u
ρ~u dV = ρ dV. (2.60)
dt Bt Bt Dt
Substituting the expression for the rate of change of momentum in (2.60)

into the left–hand side of (2.52) yields
ZZZ ZZZ ZZZ
D~u
ρ dV = − ∇p dV + f~ dV,
Bt Dt Bt Bt
or ZZZ
D~u
ρ + ∇p − f~ dV = 0, for all t. (2.61)
Bt Dt
Assuming that the fields ρ, ~u and p are C 1 over R and for all times t, and that
the field f~ is continuous over R and for all times t, we see that the integrand in
the left–hand side of (2.61) is continuous over R and for all times t. Thus, since
(2.61) holds true for all bounded subregions, Bt , of R with smooth boundary,
we conclude that
D~u
ρ + ∇p − f~ = 0, in R, for all t,
Dt
or
D~u
ρ = −∇p + f~, in R, for all t, (2.62)
Dt
which is the differential form of the conservation of momentum principle.
Observe that the PDE in (2.62) is a vector differential equation in three
dimensions. As such, it is really a system of three first–order PDEs:
Du1 ∂p


 ρ = − + f1 ;


 Dt ∂x



 Du
2 ∂p
ρ = − + f2 ; (2.63)

 Dt ∂y



 ρ Du3 = − ∂p + f .



3
Dt ∂z
The equations in (2.18) and (2.63) constitute a system of four first–order
PDEs in the (possibly) unknown scalar fields u1 , u2 , u3 , ρ and p (the body
forces field f~ can usually be determined from the outset). Thus, in order to
have any hope for solving the system of conservation equations in (2.17) and
(2.62), we need to have at least one more relation, or equation, involving the
velocity field, ~u, the density, ρ, and the pressure, p. Another relation will be
provided by the principle of conservation of energy to be discussed in the next
section.
→
−
The expression in (2.60) holds true for any C 1 vector field G in R,
→
−
→
−
ZZZ ZZZ
d DG
ρ G dV = ρ dV,
dt Bt Bt Dt
or any C 1 scalar field g,

ZZZ ZZZ
d Dg
ρg dV = ρ dV, (2.64)
dt Bt Bt Dt
Dg
where is the material derivative of g. This is known as the Transport
Dt
Theorem. We will have opportunity to apply the transport theorem in (2.64)
in the next section.
2.1.3 Conservation of Energy in Incompressible Flow

Consider the volume of the portion of the fluid in Bt at time t (see Figure 2.1.3),
ZZZ
v(t) = dV. (2.65)
Bt
As the shape of the region Bt changes with the flow, the volume of Bt might
change also. We compute the rate at which the volume changes by first rewriting
the expression for v(t) in (2.65) as
ZZZ
v(t) = J(ϕt (x, y, z), t) dV. (2.66)
B

ZZZ
dv ∂
= [J(ϕt (x, y, z), t)] xdydz, (2.67)
dt B ∂t
where
∂
[J(ϕt (x, y, z), t) = (∇ · ~u(ϕt (x, y, z), t))J(ϕt (x, y, z), t) ,
∂t
according to (2.32). We therefore obtain from (2.67) that
ZZZ
dv
= (∇ · ~u(ϕt (x, y, z), t))J(ϕt (x, y, z), t) dxdydz,
dt B
which we can rewrite as ZZZ

dv
= ∇ · ~u dV. (2.68)
dt Bt
In an incompressible flow the volume of any portion of the fluid does not
change with time. We therefore obtain from (2.68) that
ZZZ
∇ · ~u dV = 0, for all t. (2.69)
Bt
Since the expression in (2.69) holds true for any Bt in R, it follows that, for
case in which the velocity field, ~u, is C 1 in R, the condition for the flow to be
incompressible is
∇ · ~u = 0, in R for all t. (2.70)
We show in this section that, in an ideal incompressible fluid, the kinetic energy
in the portion of the fluid in Bt is conserved.
The kinetic energy of the portion of the fluid Bt at time t is given by
ZZZ
1
E(t) = ρk~uk2 dV, (2.71)
2 Bt
where kuk2 = ~u · ~u is the square of the Euclidean norm of the velocity field ~u.
The rate of change of E in (2.71) is given by the Transport Theorem in
(2.64) to be ZZZ
dE 1 D
= ρ [k~uk2 ] dV, (2.72)
dt 2 Bt Dt
where
D D~u
[k~uk2 ] = 2~u · ,
Dt Dt
so that, in view of (2.72),
ZZZ
dE D~u
= ρ~u · dV,
dt Bt Dt
or ZZZ
dE D~u
= ~u · ρ dV. (2.73)
dt Bt Dt
Substituting the law of conservation of momentum expression for an ideal fluid
in (2.62) into the right–hand side of (2.73) then yields
ZZZ
dE
= ~u · −∇p + f~ dV,
dt Bt

ZZZ ZZZ
dE
=− ∇p · ~u dV + f~ · ~u dV. (2.74)
dt Bt Bt
The right–most integral in (2.74) measures the rate at which body forces do
work in the portion of fluid in Bt at time t. In order to understate the other
integral in (2.74) we use the assumption that the fluid is incompressible, stated
as the PDE in (2.70), to obtain
∇ · (p~u) = ∇p · ~u + p∇ · ~u = ∇p · ~u,
so that ZZZ ZZZ

∇p · ~u dV = ∇ · (p~u) dV. (2.75)
Bt Bt
Applying the divergence theorem to the integral on the right–hand side of (2.75)
yields ZZZ ZZ
∇p · ~u dV = p~u · ~n dA, (2.76)
Bt ∂Bt
where ~n denotes the outward unit normal vector the boundary of Bt . Substi-
tuting the expression in (2.76) into the right–hand side of (2.74) then yields
ZZ ZZZ
dE
=− p~u · ~n dA + f~ · ~u dV. (2.77)
dt ∂Bt Bt
ZZ
Observe that − p~u · ~n dA gives the rate at which the stress forces are
∂Bt
doing work on the portion of fluid in Bt . Hence, the equation in (2.77) is a
statement of the conservation of kinetic energy.
2.1.4 Euler Equations for Incompressible, Ideal Fluids

Putting together the PDEs in (2.58), (2.62) and (2.70) we obtain the system of
PDEs 
Dρ

 = 0;
Dt






D~u (2.78)
ρ = −∇p + f~;
 Dt






∇ · ~u = 0,

stating the principles of conservation of mass, conservation of momentum, and

conservation of energy, respectively, for incompressible, ideal fluids. The equa-
tions in the system of PDEs in (2.78) are known as the Euler equations for
incompressible, ideal fluids. Using the definition of the material derivative,
D
, in (2.38) and (2.39), the Euler equations in (2.78) can also be written as
Dt

∂ρ
+ ~u · ∇ρ = 0;


∂t






∂~u (2.79)
ρ + ρ(~u · ∇)~u = −∇p + f~;
 ∂t






∇ · ~u = 0,

The fields ρ, ~u and p in (2.79) are assumed to be C 1 functions defined in and

open region R in R3 , and for t > 0; the field f~ is assumed to be continuous in
R and for all t > 0. The field f~ is usually know; but the functions ρ, ~u and p
are unknown. We would like to obtain information about these functions for all
times, t, and all points in R, given some initial conditions; for example,

 ρ(x, y, z, 0) = ρo (x, y, z), for (x, y, z) ∈ R;
~u(x, y, z, 0) = ~uo (x, y, z), for (x, y, z) ∈ R;
p(x, y, z, 0) = po (x, y, z), , for (x, y, z) ∈ R,

where ρo , ~uo and po are given functions defined in R. Since, we want the flow
to remain within the region R, we also impose the boundary condition
~u · ~n = 0, on ∂R, for all t, (2.80)
where we are assuming that R has a smooth boundary ∂R. The condition in
(2.80) forbids fluid to cross in or out of the boundary.
2.2 Modeling Diffusion

The random migration of small particles (e.g., pollen grains, large molecules,
etc.) immersed in a stationary fluid is known as diffusion. This process,
also known as Brownian motion, is caused by the random bombardment of the
particles by the fluid molecules because of thermal excitation. Brownian motion
can be modeled probabilistically by looking at motions of large ensemble of
particles. This is a microscopic view. In this section we would like to provide a
macroscopic model of diffusion based on a conservation principle.
Imagine that a certain number of Brownian particles moves within a region
→
−
R in R3 pictured in Figure 2.2.4. Assume that there is a vector field J that
Figure 2.2.4: Brownian Particles in a Region R

2.2. MODELING DIFFUSION 25
gives a measure of the number of particles that cross a unit cross–sectional at

point (x, y, z) ∈ R and time t as follows
→
−
J (x, y, z, t) · ~ndA
gives, approximately, the number of particles that cross a small section of the
surface of area dA, per unit time, in a direction perpendicular to the surface at
that point. It then follows that the number of particles per unit time crossing
the smooth boundary of a region B ⊂ R into that region (see Figure 2.2.4) is
given by
→
−
ZZ
− J (x, y, z, t) · ~n dA, (2.81)
∂B
where the minus sign in (2.81) takes into account that we are taking ~n to be
the outward unit normal to ∂B. The expression in (2.81) is called the flux of
particles across the boundary of B.
Assume that the concentration of particles in the region R at any time t is
given by a C 1 scalar field, u, so that number of particles contained in the region
B is given at time t is given by
ZZZ
NB (t) = u(x, y, x, t) dxdydx, for all t. (2.82)
B
Assuming that particles are not being created or destroyed, we get the conser-
vation principle
→
−
ZZ
dNB
=− J (x, y, z, t) · ~n dA (2.83)
dt ∂B
Since we are assuming that u is a C 1 field, we can differentiate under the integral
sign in (2.82) to rewrite (2.83) as
→
−
ZZZ ZZ
∂u
dxdydx = − J (x, y, z, t) · ~n dA (2.84)
B ∂t ∂B
→
−
If we also assume that the vector field J is a C 1 function, we can use the
Divergence Theorem to rewrite the right–hand side of (2.84) to obtain
→
−
ZZZ ZZZ
∂u
dV = − ∇ · J dV
B ∂t B
or ZZZ
∂u →
−
+∇· J dV = 0. (2.85)
B ∂t
Since (2.85) holds true for all bounded subsets, B, of R, and all times t, we
obtain the PDE
∂u →
−
+ ∇ · J = 0, in R, for all t. (2.86)
∂t
The PDE in (2.86) has two unknown functions: the concentration u and the
→
−
flux field J . Thus, in order to complete the modeling, we need a constitutive
→
−
equation relating u and J . This is provided by Fick’s First Law of Diffusion
(see [Ber83, pg. 18]):
→
−
J = −D∇ · u, in R, for all t, (2.87)
where D is a proportionality constant known as the diffusion constant of the

medium in which the particles are, or diffusivity. Observe that D in (2.87)
has units of squared length per time. The expression in (2.87) postulates that
the flux of Brownian particles is proportional to the negative gradient of the
concentration. Thus, the diffusing particles will move from regions of high
concentration to regions of low concentration.
→
−
Substituting the expression for J in (2.87) into the conservation equation
in (2.86) we obtain
∂u
− D∇ · (∇u) = 0, in R, for all t, (2.88)
∂t
where we have used the assumption that D is constant.
Assuming that u is also a C 2 function, we can use the definitions of gradient
and divergence to compute
∂2u ∂2u ∂2u
∇ · (∇u) = + 2 + 2. (2.89)
∂x2 ∂y ∂z
The expression on the right–hand side of (2.89) is known as the Laplacian of
u, and is usually denoted by the symbol ∆u, so that
∂2u ∂2u ∂2u
∆u = + 2 + 2. (2.90)
∂x2 ∂y ∂z
Another notation for ∆u found in various textbooks is ∇2 u.
In view of (2.89) and (2.90), we see that the PDE in (2.88) can be written
as
∂u
= D∆u, in R, for all t, (2.91)
∂t
which is called the diffusion equation. The expression in (2.91) is also known
as Fick’s second equation (see [Ber83, pg. 20]), or Fick’s Second Law of Diffu-
sion.
For the case of in which the diffusing substance is constrained to move in
one space direction (say, parallel to the x–axis), the diffusion equation in (2.91)
becomes
∂u ∂2u
= D 2. (2.92)
∂t ∂x
The equation in (2.92) applies to the situation in which medium containing
Brownian particles is in a cylindrical region of constant cross sectional area and
axis parallel to the x–axis. In later chapter in these notes, we will show how to
solve the PDE in (2.92) over the entire real line subject to an initial condition
u(x, 0) = f (x), for all x ∈ R,

2.3. VARIATIONAL PROBLEMS 27
∂u
for some given function f : R → R, and some integrability conditions on u,
∂x
and f .
The equation in (2.92) also describes the flow of heat in a cylindrical metal
rod of constant cross–sectional area whose cylindrical boundary is insulated so
that heat can only flow in or out of the rod through the cross sections a the ends
of the rod (see Assignment #4). In this case u(x, t) denotes the temperature in
x
0 L
Figure 2.2.5: Heat Conduction in a Cylindrical Rod
the cross–section of the rod located at x and at time t, and the constant D is
given by
κ
D= ,
cρ
where ρ is the density, c is the specific heat, and κ is the heat conductivity of
the material the of the rod (see Assignment #4). Thus, (2.92) ia also called the
heat equation. In this case D is called the thermal diffusivity.
In these notes we will see how to solve the heat equation in (2.92) subject
the initial and boundary conditions

 u(x, 0) = f (x), for 0 < x < L;
u(0, t) = To (t), for t > 0;
u(L, t) = TL (t), for t > 0,

where f , To and TL are given functions of single variable. We will also solve the
problem with the boundary conditions

 ∂u (0, t) = 0,

for t > 0;
∂x
∂u

 (L, t) = 0, for t > 0.
∂x
These conditions imply that heat cannot flow through the end cross–sections
either; so that the rod is totally insulated.
2.3 Variational Problems

In the previous two sections we have seen how conservation principles give rise
to problems involving PDEs. Another important source of PDE problems arises
from the application of variational principles. A variational principle states
that a configuration, or function, describing the state of a system must minimize,
or maximize, certain quantity (e.g., energy). In this section we will see two
applications of variational principles: the derivations of the minimal surface
equation and the vibrating string equation.
2.3.1 Minimal Surfaces

Imagine you take a twisted wire loop, as that pictured in Figure 2.3.6, and dip
into a soap solution. When you pull it out of the solution, a soap film spanning
the wire loop develops. We are interested in understanding the mathematical
properties of the film, which can be modeled by a smooth surface in three
x R
Figure 2.3.6: Wire Loop
dimensional space. Specifically, the shape of the soap film spanning the wire
loop, can be modeled by the graph of a smooth function, u : R → R, defined on
the closure of a bounded region, R, in the xy–plane with smooth boundary ∂R.
The physical explanation for the shape of the soap film relies on the variational
principle that states that, at equilibrium, the configuration of the film must be
such that the energy associated with the surface tension in the film must be the
lowest possible. Since the energy associated with surface tension in the film is
proportional to the area of the surface, it follows from the least–energy principle
that a soap film must minimize the area; in other words, the soap film spanning
the wire loop must have the shape of a smooth surface in space containing
the wire loop with the property that it has the smallest possible area among
all smooth surfaces that span the wire loop. In this section we will develop a
mathematical formulation of this variational problem.
The wire loop can be modeled by the curve determined by the set of points:
(x, y, g(x, y)), for (x, y) ∈ ∂R,
where ∂R is the smooth boundary of a bounded open region R in the xy–plane

(see Figure 2.3.6), and g is a given function defined in a neighborhood of ∂R,
which is assumed to be continuous. A surface, S, spanning the wire loop can

be modeled by the image of a C 1 map
Φ : R → R3
given by
Φ(x, y) = (x, y, u(x, u)), for all x ∈ R, (2.93)
where R = R ∪ ∂R is the closure of R, and
u: R → R
is a function that is assumed to be C 2 in R and continuous on R; we write
u ∈ C 2 (R) ∩ C(R).
Let Ag denote the collection of functions u ∈ C 2 (R) ∩ C(R) satisfying
u(x, y) = g(x, y), for all (x, y) ∈ ∂R;
that is,
Ag = {u ∈ C 2 (R) ∩ C(R) | u = g on ∂R}. (2.94)
Next, we see how to compute the area of the surface Su = Φ(R), where Φ is
the map given in (2.93) for u ∈ Ag , where Ag is the class of functions defined
in (2.94).
The grid lines x = c and y = d, for arbitrary constants c and d, are mapped
by the parametrization Φ into curves in the surface Su given by
y 7→ Φ(c, y)
and
x 7→ Φ(x, d),
respectively. The tangent vectors to these paths are given by

∂u
Φy = 0, 1, (2.95)
∂y
and
∂u
Φx = 1, 0, , (2.96)
∂x
respectively. The quantity
kΦx × Φy k∆x∆y (2.97)
gives an approximation to the area of portion of the surface Su that results
from mapping the rectangle [x, x + ∆x] × [y, y + ∆y] in the region R to the
surface Su by means of the parametrization Φ given in (2.93). Adding up all
the contributions in (2.97), while refining the grid, yields the following formula
for the area Su : ZZ
area(Su ) = kΦx × Φy k dxdy. (2.98)
R
Using the definitions of the tangent vectors Φx and Φy in (2.95) and (2.96),
respectively, we obtain that

∂u ∂u
Φx × Φ y = − , − , 1 ,
∂x ∂y
so that s 2 2
∂u ∂u
kΦx × Φy k = 1+ + ,
∂x ∂y
or p
kΦx × Φy k = 1 + |∇u|2 ,
where |∇u| denotes the Euclidean norm of ∇u. We can therefore write (2.98)
as ZZ p
area(Su ) = 1 + |∇u|2 dxdy. (2.99)
R
The formula in (2.99) allows us to define a map
A : Ag → R
by ZZ p
A(u) = 1 + |∇u|2 dxdy, for all u ∈ Ag , (2.100)
R
which gives the area of the surface parametrized by the map Φ : R → R3 given
in (2.93) for u ∈ Ag . We will refer to the map A : Ag → R defined in (2.100)
as the area functional. With the new notation we can restate the variational
problem of this section as follows:
Problem 2.3.1 (Variational Problem 1). Out of all functions in Ag , find one
such that
A(u) 6 A(v), for all v ∈ Ag . (2.101)
That is, find a function in Ag that minimizes the area functional in the class
Ag .
Problem 2.3.1 is an instance of what has been known as Plateau’s problem
in the Calculus of Variations. The mathematical question surrounding Pateau’s
problem was first formulated by Euler and Lagrange around 1760. In the middle
of the 19th century, the Belgian physicist Joseph Plateu conducted experiments
with soap films that led him to the conjecture that soap films that form around
wire loops are of minimal surface area. It was not until 1931 that the American
mathematician Jesse Douglas and the Hungarian mathematician Tibor Radó,
independently, came up with the first mathematical proofs for the existence of
minimal surfaces. In this section we will derive a necessary condition for the
existence of a solution to Problem 2.3.1, which is expressed in terms of a PDE
that u ∈ Ag must satisfy, the minimal surface equation.
Suppose we have found a solution, u ∈ Ag , to Problem 2.3.1 in u ∈ Ag . Let

ϕ : R → R by a C ∞ function with compact support in R; we write ϕ ∈ Cc∞ (R)
(see Assignment #5 for a construction of such function). It then follows that
u + tϕ ∈ Ag , for all t ∈ R, (2.102)
since ϕ vanishes in a neighborhood of ∂R and therefore u + tϕ = g on ∂R. It

follows from (2.102) and (2.101) that
A(u) 6 A(u + tϕ), for all t ∈ R. (2.103)
Consequently, the function f : R → R defined by
f (t) = A(u + tϕ), for all t ∈ R, (2.104)
has a minimum at 0, by virtue of (2.104) and (2.104). It follows from this

observation that, if f is differentiable at 0, then
f 0 (0) = 0. (2.105)
We will see next that, since we are assuming that u ∈ C 2 (R) ∩ C(R) and
ϕ ∈ Cc∞ (R), f is indeed differentiable. To see why this is the case, use (2.104)
and (2.100) to compute
ZZ p
f (t) = 1 + |∇(u + tϕ)|2 dxdy, for all t ∈ R, (2.106)
R
where
∇(u + tϕ) = ∇u + t∇ϕ, for all t ∈ R,
by the linearity of the differential operator ∇. It then follows that
∇(u + tϕ) = (∇u + t∇ϕ) · (∇u + t∇ϕ)
= ∇u · ∇u + t∇u · ∇ϕ + t∇ϕ · ∇u + t2 ∇ϕ · ∇ϕ
= |∇u|2 + 2t∇u · ∇ϕ + t2 |∇ϕ|2 ,
so that, substituting into (2.106),

ZZ p
f (t) = 1 + |∇u|2 + 2t∇u · ∇ϕ + t2 |∇ϕ|2 dxdy, for all t ∈ R. (2.107)
R
Since the integrand in (2.107) is C 1 , we can differentiate under the integral sign
to get
∇u · ∇ϕ + t|∇ϕ|2
ZZ
0
f (t) = p dxdy, (2.108)
R 1 + |∇u|2 + 2t∇u · ∇ϕ + t2 |∇ϕ|2
for all t ∈ R. Thus, f is differentiable and, substituting 0 for t in (2.108),

∇u · ∇ϕ
ZZ
f 0 (0) = p dxdy. (2.109)
R 1 + |∇u|2
Hence, if u is a minimizer of the area functional in Ag , it follows from (2.104)

and (2.109) that
∇u · ∇ϕ
ZZ
p dxdy = 0, for all ϕ ∈ Cc∞ (R). (2.110)
R 1 + |∇u|2
The statement in (2.110) provides a necessary condition for the existence of

a minimizer of the area functional in Ag . We will next see how (2.110) gives rise
to a PDE that u ∈ C 2 (R) ∩ C(R) must satisfy in order for it to be minimizer
of the area functional in Ag .
First, we “integrate by parts” (see Assignment #6) in (2.110) to get
!
∇u ∇u · ~n
ZZ Z
− ∇· p ϕ dxdy + ϕp ds = 0, (2.111)
1 + |∇u| 2 1 + |∇u|2
R ∂R
for all ϕ ∈ Cc∞ (R), where the second integral in (2.111) is a path integral
around the boundary of R. Since ϕ ∈ Cc∞ (R) vanishes in a neighborhood of the
boundary of R, it follows from (2.111) that
!
∇u
ZZ
∇· p ϕ dxdy = 0, for all ϕ ∈ Cc∞ (R). (2.112)
1 + |∇u| 2
R
By virtue of the assumption that u is a C 2 functions, it follows that the di-

vergence term of the integrand (2.112) is continuous on R, it follows from the
statement in (2.112) that
!
∇u
∇· p = 0, in R. (2.113)
1 + |∇u|2
(see Assignment #6).

The equation in (2.113) is a second order nonlinear PDE known as the
minimal surface equation. It provides a necessary condition for a function
u ∈ C 2 (R) ∩ C(R) to be a minimizer of the area functional in Ag . Since, we are
also assuming that u ∈ Ag , we get that must solve the boundary value problem
(BVP):  !
 ∇u
 ∇· p = 0 in R;


1 + |∇u|2 (2.114)



u = g on ∂R.

The BVP in (2.114) is called the Dirichlet problem for the minimal surface
equation.
The PDE in (2.113) can also be written as

(1 + u2y )uxx − 2ux uy uxy + (1 + u2x )uyy = 0, in R, (2.115)
where the subscripted symbols read as follows:
∂u ∂u
ux = , uy = ,
∂x ∂y
∂2u ∂2u
uxx = , uyy = ,
∂x2 ∂y 2
and
∂2u ∂2u
uxy = = = uyx . (2.116)
∂y∂x ∂x∂y
The fact that the “mixed” second partial derivatives in (2.116) are equal follows
from the assumption that u is a C 2 function.
When we study the classification of PDEs we will see that the equation in
(2.115) is a nonlinear, second order, elliptic PDE.
2.3.2 The Linearized Minimal Surface Equation

For the case in which the wire loop in the previous section is very close to a
horizontal plane (see Figure 2.3.7), it is reasonable to assume that, if u ∈ Ag ,
x R
Figure 2.3.7: Almost Planar Wire Loop
|∇u| is very small throughout R. We can therefore use the linear approximation
√ 1
1 + t ≈ 1 + t, for small |t|, (2.117)
2
to approximate the area function in (2.100) by

ZZ
1 2
A(u) ≈ 1 + |∇u| dxdy, for all u ∈ Ag ,
R 2
so that
ZZ
1
A(u) ≈ area(R) + |∇u|2 dxdy, for all u ∈ Ag . (2.118)
2 R
The integral on the right–hand side of the expression in (2.118) is known as

the Dirichlet Integral. We will use it in these notes to define the Dirichlet
functional, D : Ag → R,
ZZ
1
D(u) = |∇u|2 dxdy, for all u ∈ Ag . (2.119)
2 R
Thus, in view of (2.118) and (2.119),
A(u) ≈ area(R) + D(u), for all u ∈ Ag . (2.120)
Thus, according to (2.120), for wire loops close to a horizontal plane, mini-
mal surfaces spanning the wire loop can be approximated by solutions to the
following variational problem,
Problem 2.3.2 (Variational Problem 2). Out of all functions in Ag , find one
such that
D(u) 6 D(v), for all v ∈ Ag . (2.121)
It can be shown that a necessary condition for u ∈ Ag to be a solution to
the Variational Problem 2.3.2 is that u solves the boundary value problem

 ∆u = 0 in R;
(2.122)
u = g on ∂R,

where
∆u = uxx + uyy ,
the two–dimensional Laplacian. The BVP in (2.122) is called the Dirichlet
Problem for Laplace’s equation.
2.3.3 Vibrating String

Consider a string of length L (imagine a guitar string or a violin string) whose
ends are located at x = 0 and x = L along the x–axis (see Figure 2.3.8). We
assume that the string is made of some material of density ρ(x) (in units of
mass per length). Assume that the string is fixed at the end–points and is
tightly stretched so that there is a constant tension, τ , acting tangentially along
the string at all times. We would like to model what happens to the string
x
0 L
Figure 2.3.8: String of Length L at Equilibrium
x
0 L
Figure 2.3.9: Plucked String of Length L
after it is plucked to a configuration like that pictured in Figure 2.3.9 and then
released. We assume that the shape of the plucked string is described by a
continuous function, f , of x, for x ∈ [0, L]. At any time t > 0, the shape of
the string is described by a function, u, of x and t; so that u(x, t) gives the
vertical displacement of a point in the string located at x when the string is in
the equilibrium position pictured in Figure 2.3.8, and at time t > 0. We then
have that
u(x, 0) = f (x), for all x ∈ [0, L]. (2.123)
In addition to the initial condition in (2.123), we will also prescribe the initial
speed of the string,
∂u
(x, 0) = g(x), for all x ∈ [0, L], (2.124)
∂t
where g is a continuous function of x; for instance, if the plucked string is
released from rest, then g(x) = 0 for all x ∈ [0, L]. We also have the boundary
conditions,
u(0, t) = u(L, t) = 0, for all t, (2.125)
which model the assumption that the ends of the string do not move.
The question we would like to answer is: Given the initial conditions in
(2.123) and (2.124), and the boundary conditions in (2.125), can we determine
the shape of the string, u(x, t), for all x ∈ [0, L] and all times t > 0. We will
answer this questions in a subsequent chapter in these notes. In this section,
though, we will derive a necessary condition in the form of a PDE that u must
satisfy in order for it to describe the motion of the vibrating string.
In order to find the PDE governing the motion of the string, we will formulate
the problem as a variational problem. We will use Hamilton’s Principle in
Mechanics, or the Principle of Least Action. This principle states that the
the path that configurations of a mechanical system take from time t = 0 to
t = T is such that a quantity called the action is minimized (or optimized)
along the path. The action is defined by
Z T
A= [K(t) − V (t)] dt, (2.126)
0
where K(t) denotes the kinetic energy of the system at time t, and V (t) its
potential energy at time t. For the case of a string whose motion is described by
small vertical displacements u(x, t), for all x ∈ [0, L] and all times t, the kinetic
energy is given by
Z L 2
1 ∂u
K(t) = ρ(x) (x, t) dx. (2.127)
2 0 ∂t
To see how (2.127) comes about, note that the kinetic energy of a particle of
mass m is
1
K = mv 2 ,
2
where v is the speed of the particle. Thus, for a small element of the string whose
projection on the x–axis is the interval [x, x+∆x], so that its approximate length
is ∆x, the kinetic energy is, approximately,
2
1 ∂u
∆K ≈ ρ(x) (x, t) . (2.128)
2 ∂t
Thus, adding up the kinetic energies in (2.128) over all elements of the string,
and letting ∆x → 0, yields the expression in (2.127), which we rewrite as
Z L
1
K(t) = ρu2t dx, for all t, (2.129)
2 0
where ut denotes the partial derivative of u with respect to t.

In order compute the potential energy of the string, we compute the work
done by the tension, τ , along the string in stretching the string from its equi-
librium length of L, to the length at time t given by
Z L p
1 + u2x dx; (2.130)
0
so that "Z #
L p
V (t) = τ 1 + u2x dx − L , for all t. (2.131)
0
Since we are considering small vertical displacements of the string, we can lin-
earize the expression in in (2.130) by means of the linear approximation in
(2.117) to get
Z L Z L Z L
p 1 1
1 + u2x dx ≈ [1 + u2x ] dx = L + u2x dx,
0 0 2 2 0
so that, substituting into (2.131),

Z L
1
V (t) ≈ τ u2x dx, for all t. (2.132)
2 0
Thus, in view of (2.126), we consider the problem of optimizing the quantity

Z T Z L
1 2 1 2
A(u) = ρu − τ u dxdt, (2.133)
0 0 2 t 2 x
where we have substitute the expressions for K(t) and V (t) in (2.129) and
(2.132), respectively, into the expression for the action in (2.126).
We will use the expression for the action in (2.133) define a functional in the
class of functions A defined as follows: Let R = (0, L) × (0, T ), the cartesian
product of the open intervals (0, L) and (0, T ). Then, R is an open rectangle
in the xt–plane. We say that u ∈ A if u ∈ C 2 (R) ∩ C(R), and u satisfies the
initial conditions in (2.123) and (2.124), and the boundary conditions in (2.125).
Then, the action functional,
A : A → R,
is defined by the expression in (2.133), so that
ZZ
1 2
ρut − τ u2x dxdt,

A(u) = for u ∈ A. (2.134)
2 R
Next, for ϕ ∈ Cc∞ (R), note that u + sϕ ∈ A, since ϕ has compact support
in R, and therefore ϕ and all its derivatives are 0 on ∂R. We can then define a
real valued function h : R → R by
h(s) = A(u + sϕ), for s ∈ R, (2.135)
Using the definition of the functional A in (2.134), we can rewrite h(s) in (2.135)
as ZZ
1
ρ[(u + sϕ)t ]2 − τ [(u + sϕ)x ]2 dxdt

h(s) =
2 R
ZZ
1
ρ[ut + sϕt ]2 − τ [ux + sϕx ]2 dxdt,

=
2 R
so that
ZZ
h(s) = A(u) + s [ρut ϕt − τ ux ϕx ] dxdt + s2 A(ϕ), (2.136)
R
for s ∈ R, where we have used the definition of the action functional in (2.134).
It follows from (2.136) that h is differentiable and
ZZ
h0 (s) = [ρut ϕt − τ ux ϕx ] dxdt + 2sA(ϕ), for s ∈ R. (2.137)
R
The principle of least action implies that, if u describes the shape of the string,
then s = 0 must be ac critical point of h. Hence, h0 (0) = 0 and (2.137) implies
that ZZ
[ρut ϕt − τ ux ϕx ] dxdt = 0, for ϕ ∈ Cc∞ (R), (2.138)
R
is a necessary condition for u(x, t) to describe the shape of a vibrating string

for all times t.
Next, we use the integration by parts formulas
ZZ Z ZZ
∂ϕ ∂ψ
ψ dxdt = ψϕn1 ds − ϕ dxdt,
R ∂x ∂R R ∂x
for C 1 functions ψ and ϕ, where n1 is the first component of the outward unit
normal, ~n, on ∂R (wherever this vector is defined), and
ZZ Z ZZ
∂ϕ ∂ψ
ψ dxdt = ψϕn2 ds − ϕ dxdt,
R ∂t ∂R R ∂t
where n2 is the second component of the outward unit normal, ~n, to obtain
ZZ Z ZZ
∂
ρut ϕt dxdt = ρut ϕn2 ds − [ρut ]ϕ dxdt,
R ∂R R ∂t
so that ZZ ZZ
∂
ρut ϕt dxdt = − [ρut ]ϕ dxdt, (2.139)
R R ∂t
since ϕ has compact support in R.
Similarly, ZZ ZZ
∂
τ ux ϕx dxdt = − [τ ux ]ϕ dxdt. (2.140)
R R ∂x
Next, substitute the results in (2.139) and (2.140) into (2.138) to get
ZZ
∂ ∂
[ρut ] − [τ ux ] ϕ dxdt = 0, for ϕ ∈ Cc∞ (R). (2.141)
R ∂t ∂x
Thus, we obtain from (2.141) that
∂2u ∂2u
ρ − τ = 0, in R, (2.142)
∂t2 ∂x2
since we area assuming that u is C 2 , ρ is a continuous function of x, and τ is
constant.
The PDE in (2.142) is called the one–dimensional wave equation. It is
sometimes written as
∂2u τ ∂2u
= ,
∂t2 ρ ∂x2
or
∂2u ∂2u
2
= c2 2 , (2.143)
∂t ∂x
where
τ
c2 = ,
ρ
for the case in which ρ and τ are assumed to be constant.
The wave equation in (2.142) or (2.143) is a second order, linear, hyperbolic
PDE.
2.4. MODELING SMALL AMPLITUDE VIBRATIONS 39
2.4 Modeling Small Amplitude Vibrations

In this section we present another derivation of the wave equation in (2.143)
based on an application of Newton’s laws of motion.
x
0 x x + ∆x L
Figure 2.4.10: Vibrating String in Motion
Figure 2.4.10 shows a snapshot of the vibrating string at some time t. The
figure also highlights a small section of the string that is above an interval
[x, x + ∆x], for some x ∈ (0, L). As in the derivation in Section 2.3.3, u(x, t) is
the vertical displacement of a point on the string located at x when the string is
at equilibrium, and time t (we are assuming here that the vibrations have small
amplitude; so that, we may assume that each particle along the string moves
along a vertical direction); ρ is the density of the material making up the string
(in units of mass per unit of length); and τ is the tension along the string in
units of force. We assume that u has continuous partial derivatives,
∂u ∂u ∂2u ∂2u ∂2u

, , , , ,
∂x ∂t ∂x2 ∂x∂t ∂t2
for all x ∈ (0, L) and all t > 0. We introduce the angle θ(x, t) that the tangent
line to the curve in Figure 2.4.10 at the point (x, u(x)) makes with the positive
x–axis and note that
∂u
tan(θ(x, t)) = (x, t), for x ∈ (0, L), and t > 0,
∂x
which we can also write more succinctly as
tan(θ) = ux . (2.144)
It follows from (2.144) that θ has continuous partial derivatives as long as

sec2 θ 6= 0, and
∂θ uxx
= ,
∂x sec2 θ
which we can rewrite as
∂θ uxx
= , (2.145)
∂x 1 + u2x
where we have used (2.144) again. We also obtain from (2.144) that
1
cos(θ) = p . (2.146)
1 + u2x
If we assume that ∆x is very small, we may treat the section of the string
above the interval [x, x + ∆x] as a particle of mass given approximately by
ρ(x)∆s,
where ∆s is the arc–length of the section of the string, which is given, approxi-
mately, by p
∆s ≈ 1 + u2x ∆x.
Thus, the mass of the section of the string above the interval [x, x + ∆x] is,
approximately, p
∆m ≈ ρ(x) 1 + u2x ∆x. (2.147)
Newton’s Second Law of Motion applied to a particle of mass ∆m given in
(2.147) reads
∂2u
(∆m) = forces acting on the particle in vertical dirction. (2.148)
∂t2
The forces acting on the particle in the vertical direction are the force of
gravity and the vertical components of the tension, which acts tangentially to
the string and away from the section above the interval [x, x+∆x]. There might
be other forces acting on the string as well; for instance, a violin string might be
acted on by the bow when playing. We will put together the forces other than
tension into a function, F (x, t), in units of force per length. Thus, the forces,
other than tension, acting on the particle are
F (x, t)∆x. (2.149)
The vertical component of the tension acting on the section of the string above
the interval [x, x + ∆x] is
τ sin(θ(x + ∆x, t)) − τ sin(θ(x, t)), (2.150)
where we are assuming that τ is constant throughout the string.

Putting together (2.147), (2.148), (2.149) and (2.150),
p ∂2u
ρ(x) 1 + u2x ∆x 2 = τ [sin(θ(x + ∆x, t)) − sin(θ(x, t))] + F (x, t)∆x. (2.151)
∂t
Next, divide the equation in (2.151) on both sides by ∆x 6= 0, and letting
∆x → 0, we obtain that
p ∂2u ∂
ρ(x) 1 + u2x =τ [sin(θ(x, t))] + F (x, t),
∂t2 ∂x
p ∂2u ∂θ
ρ(x) 1 + u2x 2
= τ cos(θ) + F (x, t);
∂t ∂x
2.4. MODELING SMALL AMPLITUDE VIBRATIONS 41
so that, using (2.145),

p uxx
ρ 1 + u2x utt = τ cos(θ) + F (x, t),
1 + u2x
and, using (2.146),

p τ uxx
ρ 1 + u2x utt = + F (x, t). (2.152)
(1 + u2x )3/2
The PDE in (2.152), as it stands, is a nonlinear equation (we will see why this
is the case in the next section on classification of PDEs). We can approximate
the equation by a linear PDE by using the assumption that the vibrations of
the string are on very small amplitude. In the case of very small amplitude
vibrations, ux , which is related to the tangent of the angle the string makes
with a horizontal direction, is very small in magnitude. Thus, we can make the
approximation
ux ≈ 0
in (2.152) to obtain
ρutt = τ uxx + F (x, t). (2.153)
Observe that the equation in (2.153) leads to the wave equation in (2.142) for
the case F (x, t) = 0 for all x ∈ [0, L] and t > 0. The equation in (2.142) will be
called in this notes the one–dimensional, linear, homogeneous wave equation.
We will refer to the equation in (2.153) with F 6≡ 0 as the one–dimensional,
linear, non–homogeneous wave equation. In these notes we will show how to
construct solutions of those equations.
Chapter 3
How are PDEs Classified?
In the previous chapter we saw how various types of PDEs.

PDEs are classified according to order (the highest order of the derivative of
the unknown functions involved in the equation). The Euler equations for an
ideal, incompressible fluid,

∂ρ
+ ~u · ∇ρ = 0;


∂t






∂~u (3.1)
ρ + ρ(~u · ∇)~u = −∇p + f~;
∂t







∇ · ~u = 0,

are a system of first–order PDEs.

The 3–dimensional diffusion equation, or heat conduction equation,
2
∂ u ∂2u ∂2u

∂u
=D + + , (3.2)
∂t ∂x2 ∂y 2 ∂z 2
the two–dimensional Laplace’s equation,
uxx + uyy = 0, (3.3)
the one dimensional wave equations,

∂2u ∂2u
2
= c2 2 , (3.4)
∂t ∂x
and the minimal surface equation,
(1 + u2y )uxx − 2ux uy uxy + (1 + u2x )uyy = 0, (3.5)
are second order PDEs.

PDEs can further be classified as linear or nonlinear. For instance, the third
equation in the system in (3.1), and the PDEs in (3.2), (3.3) and (3.4) are linear
43
44 CHAPTER 3. CLASSIFICATION OF PDES
equations, while the first two equations in the system in (3.1) and the PDE in
(3.5) are not linear. In the next section we will discuss properties of linear
equations, and how those properties can be very helpful in the construction of
solutions, and proofs of uniqueness for some initial/boundary value problems.
The PDE in (3.5) is in a general class of equations of the form
a(x, y, u, ux , uy )uxx + b(x, y, u, ux , uy )uxy + c(x, y, u, ux , uy )uyy = d, (3.6)
for some continuous function a, b, c and d of the five variables x, y, u, ux and

uy , generally. If the coefficient functions in (3.6) depend only on x and y, we
get the general linear second order equation in two variables,
a(x, y)uxx + b(x, y)uxy + c(x, y)uyy = d(x, y). (3.7)
If the coefficient functions in (3.6) do not depend on the derivatives of the

unknown function u, we obtain the quasi–linear, second order PDE
a(x, y, u)uxx + b(x, y, u)uxy + c(x, y, u)uyy = d(x, y, u). (3.8)
In Section 3.2 we will discuss a further classification of the general second

order PDE in (3.6) based on properties of certain curves associated with the
equation known as characteristic curves. This will lead to the definition of
elliptic, hyperbolic and parabolic PDEs. Laplace’s equation,
uxx + uyy = 0, (3.9)
the one–dimensional wave equation,

1
uxx − utt = 0, (3.10)
c2
and the one–dimensional diffusion equations,
Duxx − ut = 0, (3.11)
are archetypes of these classes of equations, respectively.
3.1 Linearity
Laplace’s equation (3.9), the one–dimensional wave equation (3.10), and the
one–dimensional diffusion equations (3.11) are linear equations. To understand
the use of this terminology in the context of PDEs, note that Laplace’s equation
(3.9) can also be written as
∆u = 0,
where ∆ : C 2 (R) → C(R) defines a linear operator between the spaces of func-
tions C 2 (R) and C(R) given by
∆u = uxx + uyy , for all u ∈ C ( R), (3.12)

3.1. LINEARITY 45
for some open subset R of R2 . The differential ∆ defined in (3.12) is linear

because of the linearity property of differentiation that we learned in Calculus;
indeed, given functions u, v ∈ C 2 (R) and real constants c1 and c2 , it follows
from the linearity of differentiation that
∆(c1 u + c2 v) = c1 ∆u + c2 ∆v.
Similarly, the one–dimensional wave equation in (3.10) can be written as
−u = 0,
where the linear operator : C 2 (R) → C(R), also known as the d’Alembert
operator, is defined by
1
u = utt − uxx , for all u ∈ C 2 (R),
c2
where R is an open region in the xt–plane; and the one–dimensional diffusion
equation in (3.11) can be written as
−Lu = 0,
where L : C 2 (R) → C(R) defined by
Lu = ut − Duxx , for all u ∈ C 2 (R),
where R is an open region in the xt–plane, is also a linear operator.
By contrast, the map N : C 1 (R) × C 1 (R) × C 1 (R) → C(R) × C(R) × C(R),
given by
∂~u
N (~u) = + (~u · ∇)~u, for all ~u ∈ C 1 (R) × C 1 (R) × C 1 (R),
∂t
where R is an open set in R3 , is not linear (see Problem 1 in Assignment #9).
Hence, the second PDE in the system (3.1) is not a linear equation.
In general, a linear PDE is an expression of the form
Lu = f, (3.13)
where L : U → F is a linear differential operator from a linear space, U, of
differentiable functions to a linear space, F, of continuous functions. An example
of the equation in (3.13) is provided by Poisson’s equation in Potential Theory,
−∆u = f, in R, (3.14)
where R is an open region in Rn . In this case, the linear operator L = −∆ maps
C 2 (R) to C(R).
If f = 0 in the right–hand side of (3.13) we get the homogeneous PDE
Lu = 0. (3.15)
The equation in (3.15) has the following very useful property known as the
principle of superposition.
Proposition 3.1.1 (Principle of Superposition). Let u and v denote two so-

lutions of the homogeneous PDE in (3.15). Then, for any constants c1 and c2 ,
the functions c1 u + c2 v is also a solution of (3.15).
Proof: Since L is a linear differential operator, it follows that
L(c1 u + c2 v) = c1 Lu + c2 Lv.
Thus, if u and v solve (3.15), it follows that
L(c1 u + c2 v) = c1 0 + c2 0 = 0,
which shows that c1 u + c2 v solves (3.15).
3.2 Classification of Second Order PDEs

Laplace’s equation (3.9), the one–dimensional wave equation (3.10), and the
one–dimensional diffusion equations (3.11) are second order PDEs. They are
classified as elliptic, hyperbolic and parabolic PDEs, respectively. In this
sections we study the rationale of that classification as it applies to the general
second order PDE in two variables:
a(x, y, u, ux , uy )uxx + b(x, y, u, ux , uy )uxy + c(x, y, u, ux , uy )uyy = d. (3.16)
We begin with the special case of the linear equation
a(x, y)uxx + b(x, y)uxy + c(x, y)uyy = d(x, y), (3.17)
where a, b, c and d are continuous functions defined in some opens subset, R, of

R2 . The classification of the equations of the type in (3.16) or (3.17) is based on
properties of curves in R associated with the equations; these curves are called
characteristic curves. We begin with a curve, Γ, in R parametrized by a map
γ : I → R2 ,
t 7→ γ(t) = (x(t), y(t)), for t ∈ I,
where I is some interval of real numbers; see Figure 3.2.1. Suppose we are
trying to solve the linear PDE in (3.17) subject to information about u given
on the curve Γ. Specifically, suppose we are given the values of u and its first
derivatives on Γ; we can specify this conditions these condition on u by the
equations
u(x(t), y(t)) = uo (t), for t ∈ I, (3.18)
ux (x(t), y(t)) = f (t), for t ∈ I, (3.19)
and
uy (x(t), y(t)) = g(t), for t ∈ I, (3.20)
where uo , f and g are given continuous functions on I. If we assume, in addition,
that f and g are C ∞ functions, we can in theory obtain information about the
3.2. SECOND ORDER PDES 47
(x(t), y(t))
Figure 3.2.1: Characteristic Curves
second order derivatives, uxx , uxy and uyy , of u (and higher order derivatives)
on Γ. Is this can be done, we can attempt to construct a solution of the PDE in
(3.17) by building Taylor series expansions around every point on Γ using the
values of u and its derivatives based on the conditions in (3.18), (3.19) and (3.20)
and derivatives of the expressions in (3.19) and (3.20). The first step in this
construction consists of taking the derivatives of derivatives of the expressions
in (3.19) and (3.20) and combining these with the information provided by the
PDE (3.17) to obtain the linear system
= f˙

 ẋ uxx + ẏ uxy
ẋ uxy + ẏ uyy = ġ (3.21)
a uxx + b uxy + c uyy = d,

for the unknowns uxx , uxy and uyy on Γ, where a dot on top of a variable
denotes derivative with respect to t:
dx dy ˙ df dg
ẋ = , ẏ = , f= and ġ = .
dt dt dt dt
Note that the system in (3.21) can be written in matrix form as
f˙
    
ẋ ẏ 0 uxx
 0 ẋ ẏ  uxy  =  ġ  . (3.22)
a b c uyy d
The matrix equation in (3.22) can be solved for the second derivatives of u, in
terms of the data (f˙, ġ, d) on Γ, provided that the determinant of the matrix
 
ẋ ẏ 0
 0 ẋ ẏ  (3.23)
a b c
is not zero. The case in which the determinant of the matrix in (3.23) yields
the equation
a(ẏ)2 − bẋẏ + c(ẋ)2 = 0. (3.24)
Dividing the equation in (3.24) by ẋ2 and using the fact that
ẏ dy
= ,
ẋ dx
by the Chain Rule, we obtain the ordinary differential equation (ODE)
2
dy dy
a −b + c = 0. (3.25)
dx dx
We shall refer to the ODE in (3.25) as the characteristic equation to the PDE
in (3.17). Solutions to the ODE in (3.25) are called characteristic curves of
dy
the PDE in (3.17). Assuming that a 6= 0 in R, we can solve for in (3.25) to
dx
get √
dy b ± b2 − 4ac
= . (3.26)
dx 2a
We have three possibilities, depending on whether b2 − 4ac is positive, zero, or
negative.
If b2 − 4ac > 0, then the PDE in (3.17) has two families of characteristic
curves given by the solutions to the two ODEs in (3.26). In this case we say
that the PDE in (3.17) is hyperbolic.
If b2 −4ac = 0, then the PDE in (3.17) has one family of characteristic curves
given by the solution to the ODE
dy b
= .
dx 2a
In this case we say that the PDE in (3.17) is parabolic.
If b2 − 4ac < 0, then the ODE in (3.25) has no real solutions. Thus, the
PDE in (3.17) has no (real) characteristic curves. In this case we say that the
PDE in (3.17) is elliptic.
Example 3.2.1 (The One–dimensional Wave Equation). In the case of the
linear second order equation
c2 uxx − utt = 0, (3.27)
describing small amplitude vibrations of a string that we derived in Section 2.3.3

(see the PDE in (2.143), a = c2 , b = 0 and c in (3.25) is −1 (in this case t is
playing the role of y). We therefore get that b2 − 4ac = −4(c2 )(−1) = 4c2 > 0;
so that the equation in (3.27) is hyperbolic. For this PDE the equations for the
characteristic curves in (3.26) yields
dt 2c
= ± 2,
dx 2c
3.2. SECOND ORDER PDES 49
or
dt 1
=± ,
dx c
dx
= ±c. (3.28)
dt
The equations in (3.28) is a pair of ODEs that can be solved to yield the two
families of curves
x = ct + ξ, (3.29)
and
x = −ct + η, (3.30)
where ξ and η are the parameters for each of the families of characteristic
curves in (3.29) and (3.30). The family of characteristic curves described by
the equations in (3.29) consists of parallel lines of (positive) slope 1/c in the
xt–plane with x–intercept ξ ∈ R. Some of those characteristic curves are shown
in Figure 3.2.2.
Figure 3.2.2: Characteristic Curves of utt = c2 uxx
The family of characteristic curves described by the equations in (3.29) con-

sist of parallel lines of (negative) slope −1/c in the xt–plane and x–intercept
ηıR; some of theses curves are sketched in Figure 3.2.3.
Figure 3.2.4 contains a sketch of both sets of characteristic curve in the same
graph. We will see in Example 4.1.1 of Section 4.1 how to use the two sets of
characteristic curves in Figure 3.2.4 to construct a solution to the initial value
problem to the one–dimensional wave equation.

Chapter 4
How are PDEs Solved?
There is really no general theory for solving any given PDE of the form in (2.1),
F (x, u, ux1 , . . . , uxn , ux1 x2 , . . . , uxn xn , . . .) = 0.
Approaches to the construction of solutions of PDE problems are determined

by the types of PDEs and the geometric properties (e.g., symmetries) of the
equations and/or domains in which the problems are posed. In this chapter we
present some of those approaches. Emphasis will be placed on a few general
principles that can aid us when looking for solutions of PDEs.
We will begin with an approach that uses knowledge of the characteristic
curves of the equations. We will then look at approaches that exploit any
symmetries that the equations or domains in the problems might have. We will
then look at methods of solutions for linear equations based on the principle of
superposition.
4.1 Using Characteristic Curves to Solve PDEs
In Section 3.2 we defined characteristic curves for second order PDEs in two
variables, and saw how characteristic curves can be used to come up with a
classification scheme for those equations. In this section we see how to use
characteristic curves to construct solutions to certain types of PDEs in two
variables. We begin with the example of the one–dimensional wave equation.
51
52 CHAPTER 4. SOLVING PDES
4.1.1 Solving the One–Dimensional Wave Equation

Example 4.1.1 (Solving the One–Dimensional Wave Equation). We consider
the initial value problem in the entire real:
 1
uxx − c2 utt = 0, for x ∈ R, t > 0,




(4.1)


 u(x, 0) = f (x), for all x ∈ [0, L];

ut (x, 0) = g(x), for all x ∈ [0, L],

where f and g are given continuous functions defined in R.

In Example 3.2.1 in the previous section we showed that the PDE in (4.1)
has two families of characteristic curves given by
x = ct + ξ, (4.2)
and
x = −ct + η. (4.3)
These families of curves consist of parallel straight lines in the xt–plane of
slope 1/c and of slope −1/c, respectively. We will see in the next section that
characteristic curves carry information about solutions of the equation from one
point on the curve to another point on the same curve. This suggests that we
we consider the PDE in (4.1) along the curves given in (4.2) and (4.3). We can
do this my considering the parameters ξ and η in (4.2) and (4.3) as a new set
of variables,
ξ = x − ct, (4.4)
and
η = x + ct. (4.5)
If we are give a solution, u, to the PDE in (4.1), we can use the change of
variables provided by (4.4) and (4.5) to define a function, v, of ξ and η in terms
of u by means of
v(ξ, η) = u(x, t), (4.6)
where x and t can be obtained in terms of ξ and θ by solving the linear system

x − ct = ξ;
x + ct = η,
so that
1 1
x = η + ξ;
2 2
(4.7)
1 1
t = η − ξ.
2c 2c
Alternatively, we can rewrite (4.6) as
u(x, t) = v(ξ, η), (4.8)

4.1. USING CHARACTERISTIC CURVES 53
where ξ and η are given by (4.4) and (4.5), respectively.

Assume that u ∈ C 2 (R2 ) solves the PDE in (4.1). We would like to derive
a PDE satisfied by the function v defined in (4.6) and (4.7). The PDE that v
will satisfy will be expressed in terms of the new variables ξ and η. In order to
do this, we use (4.8) and the Chain Rule to get
∂ξ ∂η
ux = vξ + vη , (4.9)
∂x ∂x
where
∂ξ ∂η
= 1 and = 1, (4.10)
∂x ∂x
by virtue of (4.4) and (4.5). Thus, substituting the expressions in (4.10) into
(4.9),
ux = vξ + vη . (4.11)
Next, take partial derivative with respect to x on both sides of (4.11) and use
the Chain Rule to get
∂ ∂
uxx = [vξ ] + [vη ]
∂x ∂x
∂ξ ∂η ∂ξ ∂η
= vξξ + vξη + vηξ + vηη ,
∂x ∂x ∂x ∂x
so that, using the expressions in (4.10) and the fact that mixed partial deriva-
tives, vξη and vηξ , of C 2 functions are equal,
uxx = vξξ + 2vξη + vηη . (4.12)
Similar calculations to those leading to (4.12) can be used to obtain an

expression for utt . Indeed, take partial derivative with respect to t on both
sides of (4.8) and use the Chain Rule to get
∂ξ ∂η
ut = vξ + vη , (4.13)
∂t ∂t
where
∂ξ ∂η
= −c and = c, (4.14)
∂x ∂x
by virtue of (4.4) and (4.5). Thus, substituting the expressions in (4.14) into
(4.13),
ut = −cvξ + cvη . (4.15)
Next, take partial derivative with respect to t on both sides of (4.15) and use
the Chain Rule to get
∂ ∂
utt = [vξ ] + [vη ]
∂t ∂t
∂ξ ∂η ∂ξ ∂η
= −cvξξ − cvξη + cvηξ + cvηη ;
∂t ∂t ∂t ∂t
thus, using the expressions in (4.14), we get
utt = c2 vξξ − c2 vξη − c2 vηξ + c2 vηη ,
or
utt = c2 [vξξ − 2vξη + vηη ], (4.16)
where we have used the equality of the mixed second partial derivatives.
Since we are assuming that u solves the PDE
1
uxx − utt = 0, (4.17)
c2
it follows from (4.12), (4.16) and (4.17) that v solves the second order PDE
vξη = 0. (4.18)
Note that the PDE in (4.18) is also a hyperbolic, second order, linear PDE
(in this case a = c = 0 and b = 1, so that b2 − 4ac = 1 > 0). In contrasts
with the hyperbolic PDE in (4.17), the PDE in (4.18) can be solved directly by
integration. Indeed, writing (4.18) as
∂
[vξ ] = 0,
∂η
we see that
vξ = h(ξ), (4.19)
1
where h is an arbitrary C function of a single variable; and writing (4.19) as
∂
[v(ξ, η)] = 0,
∂ξ
we see that
v(ξ, η) = F (ξ) + G(η), (4.20)
where F is an antiderivative of h (i.e., F 0 = h), and G is an arbitrary C 2 function
of a single variable.
The function v defined by the expression in (4.20), where F and G are
arbitrary C 2 functions of a single variable, is the general solution to the PDE
in (4.18). We can use it, along with (4.8), (4.4) and (4.5) to obtain the general
solution to the one–dimensional wave equation
utt = c2 uxx , for x ∈ R, and t ∈ R; (4.21)
namely,
u(x, t) = F (x − ct) + G(x + ct), (4.22)
where F and G are arbitrary C 2 functions of a single variable. The expression in
(4.22) is known as d’Alembert’s solutions to the one–dimensional wave equation.
We now use the general solution (4.22) to the one–dimensional wave equation
in (4.21) to construct a solution to the initial value problem in (4.1). In this
construction we will need to assume that f is C 2 and g is C 1 .
Differentiate the expression for u in (4.22) with respect to t to obtain
ut (x, t) = −cF 0 (x − ct) + cG0 (x + ct), (4.23)
where we have applied the Chain Rule. Next, apply the initial conditions in
(4.1) to obtain the equations

F (x) + G(x) = f (x);
(4.24)
−cF 0 (x) + cG0 (x) = g(x),
for all x ∈ R, where we have used (4.22) and (4.23).

Next, differentiate the first equation in (4.24) and divide the second equation
by c to get
F 0 (x) + G0 (x) = f 0 (x);

(4.25)
−F 0 (x) + G0 (x) = g(x)/c,
for all x ∈ R, where we have used the differentiability assumptions on f .
Adding the equations in (4.25) then yields the following equation for G0 :
1
G0 (x) = f 0 (x) + g(x), for all x ∈ R. (4.26)
2c
Integrating the equation in (4.26) then yields
Z x
1 1
G(x) = f (x) + g(z) dz + C1 , for all x ∈ R, (4.27)
2 2c 0
where C1 is a constant of integration.
Similarly, subtracting the second equation in (4.25) from the first equation
and integrating yields
Z x
1 1
F (x) = f (x) − g(z) dz + C2 , for all x ∈ R, (4.28)
2 2c 0
where C2 is a constant of integration.
Next, substitute the functions in (4.28) and (4.27) into the formula for u(x, t)
in (4.22) to get
Z x+ct
1 1
u(x, t) = [f (x − ct) + f (x + ct)] + g(z) dz + C3 , (4.29)
2 2c x−ct
for all x ∈ R, where C3 = C1 + C2 .

It follows from the first initial condition in (4.1) that the constant C3 in
(4.29) must be 0, so that
Z x+ct
1 1
u(x, t) = [f (x−ct)+f (x+ct)]+ g(z) dz, for x ∈ R, t ∈ R, (4.30)
2 2c x−ct
(x, t)
ξ η x
solves the initial value problem (4.1) for the one–dimensional wave equation.
In order to understand what the solution to the IVP in (4.1) displayed in
(4.30) is saying, refer to Figure 4.1.1. Suppose we want to compute the value
of u at x and at time t > 0; that is, u(x, t), where (x, t) is a point in the
xt–plane. Two characteristic curves cross at that point: one with x–intercept
labeled ξ in Figure 4.1.1, and the other with x–intercept labeled η in the figure.
These correspond to the values x − ct and x + ct, respectively. According to the
expression for u in (4.29), the value of u at (x, t) is the average the values of the
initial data f at those two points, plus t times the average of all the values of
the initial speed, g, over the interval [ξ, η].
For the special case in which the initial speed is zero throughout R, we obtain
from (4.30) the special form
1
u(x, t) = [f (x − ct) + f (x + ct)], for x ∈ R and t ∈ R. (4.31)
2
1
The function u in (4.31) is made up of two traveling wave forms: f (x−ct),
2
1
which moves to the right with speed c, and f (x + ct), which moves to the left
2
with speed c. We illustrate this for the spacial case in which the initial data
f is in Cc∞ (R), with supp(f ) = [−1, 1]; see Figure 4.1.2. Figure 4.1.3 shows
the supports of the initial data and two of the traveling waves at some time
t > 0 later with ct > 2. Figure 4.1.4 shows the two pulses traveling in opposite
directions at that instant of time. Note that the two pulses in Figure 4.1.4 have
half of the amplitude of the initial pulse in Figure 4.1.2.
−1 x
1
Figure 4.1.2: Initial Data
−1 x
1
Figure 4.1.3: Traveling Waves
4.1.2 Solving First–Order PDEs

In this section we define characteristic curves for the first order equation in two
variables
∂u ∂u
a(x, y, u) + b(x, y, u) = c(x, y, u), (4.32)
∂x ∂y
where a, b and c are C ∞ functions of three real variables, (x, y, z), where (x, y)
lies in an open region, R, in R2 . For the case in which coefficient functions, a,
b and c, in (4.32) depend only on (x, y) ∈ R, the PDE in (4.32) turns into the
linear PDE:
∂u ∂u
a(x, y) + b(x, y) = c(x, y), for (x, y) ∈ R. (4.33)
∂x ∂y
We will first define the concept of characteristic curves for the PDE in (4.33).
The discussion here is analogous to the discussion on characteristic curves for
the second order equation in (3.17) on page 46. As in that discussion, the
−1 x
1
Figure 4.1.4: Traveling Pulses
starting point is smooth curve, Γ, in R parametrized by a map γ : I → R2 ,

t 7→ γ(t) = (x(t), y(t)), for t ∈ I,
where I is some interval of real numbers; see Figure 3.2.1. Suppose we are trying
to solve the linear PDE in (4.33) subject to an “initial” condition on the curve
Γ given by
u(x(t), y(t)) = f (t), for t ∈ I, (4.34)
where f is a known smooth function defined on I. The idea is that, given the
information in (4.34), we can use that information together with the PDE in
(4.33), to obtain the values of the derivatives, ux and uy , of u on Γ. Once
these are obtained, we can differentiate (4.34) and the PDE in (4.33) to obtain
information of the second derivatives on Γ. Since we are assuming that the
coefficients, a, b and c, and the “initial” data, f , are C ∞ functions, we can,
in theory, proceed in this fashion to obtain information about the higher order
derivatives of u on Γ. If this can be done, we can attempt to construct a solution
of the PDE in (4.33) by building Taylor series expansions around every point
on Γ using the values of u and its derivatives. The first step in this construction
is possible provided that the linear system
ẋ ux + ẏ uy = f˙

(4.35)
a ux + b uy = c,
for the unknowns ux , and uy on Γ can be solved. The system in (4.35) can be
written in matrix form as
f˙

ẋ ẏ ux
= . (4.36)
a b uy c
The matrix equation in (4.36) can be solved for the first derivatives of u, in
terms of the data f˙ on Γ, provided that the determinant of the matrix

ẋ ẏ
(4.37)
a b
is not zero. The case in which the determinant of the matrix in (4.37) is zero
yields the equation for the characteristic curves of the first order PDE in (4.33):
bẋ − aẏ = 0. (4.38)
Observe that the ODE in (4.38) is equivalent to the system of first order ODEs:

dx
 dt = a(x, y);



(4.39)

 dy

 = b(x, y).
dt
The system of ODEs in (4.39) defines the characteristic curves for the first–order
linear PDE in (4.33). Since, we are assuming that a and b ar C ∞ functions,
solutions to the system of first–order ODEs in (4.39) is guaranteed to have a
unique solution around to ∈ R subject to the the initial condition (x(to ), y(to )) =
(xo , yo ). Thus, in theory, characteristic curves for the PDE in (4.33) can always
be computed.
Suppose that we have computed the characteristic curves for the PDE in
(4.33) according to the system of ODEs in (4.39). Let one of those characteristics
be given by the parametrization
t 7→ (x(t), y(t)), for t ∈ I, (4.40)
where I is a maximal interval of existence. Suppose that u is a solution of

the PDE in (4.32) and consider the values of u on the characteristic curve
parametrized by (4.40),
u(x(t), y(t)), for t ∈ I. (4.41)
It follows from (4.41) and the Chain Rule that

d ∂u dx ∂u dy
[u(x(t), y(t))] = · + · ,
dt ∂x dt ∂y dt
so that, using the definition of the characteristic curves in (4.39),
d ∂u ∂u
[u(x(t), y(t))] = a(x, y) + b(x, y) = c(x, y),
dt ∂x ∂y
by virtue of the PDE in (4.33). We have therefore shown that if u is a solution
of the PDE in (4.33), then u must also solve the ODE
du
= c(x, y) (4.42)
dt
along the characteristic curves. This suggests a way to construct a solution to
initial value problem for the PDE in (4.33) where the initial data is given on a
curve that is not a characteristic curve. This approach is known as the method
of characteristic curves.
Suppose we want to solve the IVP:

 a(x, y)ux + b(x, y)uy = c(x, y) in R;
(4.43)
u = f on Γ,

where Γ is a curve in R that is not a characteristic curve. The method of

characteristic curves consists of, first, finding the characteristic curves of the
PDE in (4.43) by solving the system of ODEs in (4.39). Then, solve the ODE
in (4.42). We illustrate this method by solving the following IVP for the one–
dimensional convection equation.
Example 4.1.2 (One–Dimensional Convection Equation). Consider the IVP
∂u ∂u


 +c = 0 for x ∈ R and t > 0;
∂t ∂x (4.44)

u(x, 0) = f (x) for x ∈ R,

where c is a nonzero constant and f is a given C 1 function defined in R.

In this example t is playing the role of y, so that the equations for the
characteristic curves in (4.39) become the single ODE
dx
= c, (4.45)
dt
which can be solved to yield
x = ct + ξ, (4.46)
where ξ is a real parameter indexing the characteristic curves. For the case in
which c > 0 the characteristic curves for the PDE in (4.44) are straight lines of
positive slope 1/c in the xt–plane and x–intercept ξ. Some of these curves are
sketched in Figure 4.1.5. Along each characteristic curve in (4.46), a solution
Figure 4.1.5: Characteristic Curves of ut + cux = 0
to the PDE in (4.44) solves the ODE

du
= 0, (4.47)
dt
according to (4.42). Alternatively, we can obtain (4.47) by computing

d dx
[u(x(t), t)] = ux · + ut
dt dt
= ut + cux
= 0,
where we have used the Chain Rule, (4.45), and the assumption that u solves
the PDE in (4.44).
We can solve the ODE in (4.47) to obtain that
u(x, t) = constant along characteristic curves (4.48)
Since the characteristic curves in (4.46) are indexed by ξ, we can rewrite (4.48)
as
u(x, t) = F (ξ), (4.49)
where F is an arbitrary C 1 function of a real variable. Next, solve for ξ in (4.46)
and substitute into (4.49) to obtain the general solutions to the PDE in (4.44),
u(x, t) = F (x − ct), for x ∈ R and t ∈ R. (4.50)
For the case in which c > 0, (4.50) describes a traveling wave moving to the
right with speed c.
Finally, using the initial condition in (4.44), we get that
F (x) = f (x), for all x ∈ R,
so that
u(x, t) = f (x − ct), for x ∈ R and t ∈ R,
is a solution to the IVP in (4.44).
The method of characteristic curves illustrated thus far also applied to the
quasi–linear equation in (4.32). In this case, the equations to the characteristic
curves read 
dx
 dt = a(x, y, u);



(4.51)

 dy

 = b(x, y, u).
dt
Along characteristic curves u solves the ODE
du
= c(x, y, u). (4.52)
dt
In general, we might not be able to obtain an explicit formula for a solution of
the PDE in (4.32) based on the system (4.50)–(4.52). But, in some cases, we
might be able to obtain an expression that gives u(x, y) implicitly. We illustrate
this in the following example.
Example 4.1.3. Find a solution to the initial value problem

ut + uux = 0, for x ∈ R, t > 0;
(4.53)
u(x, 0) = f (x), for x ∈ R,
Here t is playing the role of y in the general discussion. In this case, the equation
for the characteristic curves is
dx
= u. (4.54)
dt
In order to solve the ODE in (4.54) we need information on the function u, which
is what ultimately we are trying too determine. The information is provided by
the differential equation that u satisfies along characteristic curves; namely,
du
= 0,
dt
which implies that u must be constant along characteristic curves. Thus, we
can set
u = F (ξ), (4.55)
where ξ is a parameter indexing the characteristic curves, and F is an arbitrary
C 1 function of a single variable. Substituting the expression for u in (4.55) into
the equation for the characteristic curves in (4.54) yields
dx
= F (ξ),
dt
which can be solved to yield the equation for the characteristic curves of the
PDE in (4.53):
x = F (ξ)t + ξ. (4.56)
Observe that in this case the characteristic curves are straight lines in the xt–
plane with x–intercept ξ and slope 1/F (ξ). Note that the slopes of the charac-
teristic curves depend on the value of the solution on the characteristic curves,
according to (4.55).
We can solve for ξ in (4.56) and use (4.55) to get
ξ = x − u(x, t)t
and then substitute this value into (4.55) to obtain an implicit formula for
u(x, t):
u(x, t) = F (x − u(x, t)t), for x ∈ R and t > 0. (4.57)
Next, use the initial condition in (4.53) to obtain from (4.57) that
F (x) = f (x), for all x ∈ R,
so that
u(x, t) = f (x − u(x, t)t), for x ∈ R and t > 0,
provides an expression that defines u(x, t) implicitly.
In the remainder of this section, we present more examples on the use of

characteristic curves to solve first order PDEs.
Example 4.1.4. Find a solution to the initial value problem

ut + ux = u, for x ∈ R, t > 0;
(4.58)
u(x, 0) = f (x), for x ∈ R,
where f is a given C 1 function.

The equation for the characteristic curves in this example is
dx
= 1,
dt
which can be solved to yield
x = t + ξ. (4.59)
Now, along characteristic curves, u solves the ODE
du
= u;
dt
so that
u = F (ξ)et , (4.60)
where F is a C 1 function of a real variable, and ξ is the parameter indexing the
characteristic curves in (4.59).
Next, solve for ξ in (4.59) and substitute into (4.60) to get the general
solution,
u(x, t) = F (x − t)et , for x ∈ R and t > 0, (4.61)
for the PDE in (4.58), where F is an arbitrary C 1 function. The initial condition
in (4.58) can now be used to obtain from (4.61) that
F (x) = f (x), for all x ∈ R.
It then follows from (4.61) that
u(x, t) = f (x − t)et , for x ∈ R and t > 0,
solves the initial value problem in (4.58).

Example 4.1.5. Find the general solution to the linear partial differential
equation
∂u ∂u
x +y = 2u, for (x, y) ∈ R2 . (4.62)
∂x ∂y
The equations for the characteristic curves are

dx
 dt = x;



(4.63)

 dy

 = y.
dt
Using the Chain Rule, we obtain from (4.63) the ODE

dy y
= , for x 6= 0. (4.64)
dx x
The ODE in (4.64) can be solved by separating variables to yield
y = ξx, (4.65)
where ξ is a real parameter. Thus, the characteristic curves for the PDE in
(4.62) is a pencil of straight lines through the origin in R2 .
Now, along the characteristic curves for the PDE in (4.62), u solves the ODE
du
= 2u. (4.66)
dt
Next, combine the ODE in (4.66) with the first ODE in (4.63) to obtain the
ODE
du 2u
= , for x 6= 0. (4.67)
dx x
The ODE in (4.67) can be solved by separation of variables to yield
u = F (ξ)x2 , (4.68)
where F is an arbitrary C 1 function, and ξ is the parameter indexing the char-

acteristic curves in (4.65).
Solving for ξ in (4.65) and substituting into (4.68) then yields the general
solution, y
u(x, y) = F x2 , for x 6= 0.
x
4.2 Using Symmetry to Solve PDEs

A partial differential equation is said to be invariant under a group of transfor-
mations if its form does not change after a changing variables according to the
transformations in the group. We illustrate this idea by looking at symmetric
solutions to Laplace’s equation in R2 .
4.2.1 Radially Symmetric Solutions to Laplace’s Equation

Suppose that u is a C 2 solution of Laplace’s equation in R2 ,
uxx + uyy = 0. (4.69)
We consider what happens

to the equation in (4.69) when we change to a new
ξ
set of variables, , given by a one–parameter group of rotations given by
η
the matrices
cos φ − sin φ
Mφ = ; (4.70)
sin φ cos φ
4.2. USING SYMMETRY TO SOLVE PDES 65
that is, rotations in the counterclockwise sense by an angle φ. We set

ξ x
= Mφ , (4.71)
η y
or
ξ = x cos φ − y sin φ;
(4.72)
η = x sin φ + y cos φ,
in view of (4.70) and (4.71). The equations in (4.72) can be solved for x and y
in terms of ξ and η by inverting the matrix in (4.70),

−1 cos φ sin φ
Mφ = M−φ = ,
− sin φ cos φ
so that
x = ξ cos φ + η sin φ;
(4.73)
y = −ξ sin φ + η cos φ.
In view of the equations in (4.73), we can think of u as a function of ξ and η,
which we will denote by v(ξ, η); so that
v(ξ, η) = u(x, y), (4.74)
where x and y on the right–hand side of (4.74) are given in terms of ξ and η in
(4.73).
Applying the Chain Rule, we obtain from (4.74) that
∂ξ ∂η
ux = vξ + vη ,
∂x ∂x
where
∂ξ ∂η
= cos φ and = sin φ, (4.75)
∂x ∂x
in view of the equations in (4.72). Thus,
ux = cos φ vξ + sin φ vη . (4.76)
Similar calculations using (4.74) and (4.72) yield
uy = − sin φ vξ + cos φ vη . (4.77)
Next, differentiate on both sides of (4.76) with respect to x and apply the Chain
Rule to get

∂ξ ∂η ∂ξ ∂η
uxx = cos φ vξξ + vξη + sin φ vηξ + vηη ,
∂x ∂x ∂x ∂x
so that, using (4.75) and the fact that the mixed second partial derivatives of
C 2 functions are equal,
uxx = cos2 φ vξξ + 2 sin φ cos φ vξη + sin2 φ vηη . (4.78)

Similarly, taking the partial derivative with respect to y on both sides of (4.77),
and using
∂ξ ∂η
= − sin φ and = cos φ,
∂y ∂y
which follow from (4.72), we obtain from (4.77) that
uyy = sin2 φ vξξ − 2 sin φ cos φ vξη + cos2 φ vηη . (4.79)
Thus, adding the expressions in (4.78) and (4.79),
uxx + uyy = vξξ + vηη .
Hence, if u solves Laplace’s equation in (4.69), then v solves the equation
vξξ + vηη = 0,
which has the same form as Laplace’s equation. We therefore conclude that
Laplace’s equation in (4.69) is invariant under rotations. This suggests that we
look for solutions of (4.69) that are functions of a combination of the independent
variables that is independent of the rotation parameter φ. To obtain such a
combination, use (4.72) to compute
ξ2 + η2 = (x cos φ − y sin φ)2 + (x sin φ + y cos φ)2
= x2 cos2 φ − 2xy cos φ sin φ + y 2 sin2 φ

+x2 sin2 φ + 2xy sin φ cos φ + y 2 cos2 φ
= x2 + y 2 ,
p
so that x2 +y 2 or x2 + y 2 are combinations of the independent variables, x and
y, that do not depend on φ, the rotation parameter; that is, they are rotationally
invariant. We will therefore look for solutions of the Laplace’s equation in (4.69)
that are of the form
p
u(x, y) = f ( x2 + y 2 ), for (x, y) ∈ R2 , (4.80)
where f is a C 2 function of a single variable. Functions of the form in (4.80)

are said to be radially symmetric.
Example 4.2.1 (Radially Symmetric Solutions of Laplace’s Equation in R2 ).
Let Ω = {(x, y) ∈ R2 | (x, y) 6= (0, 0)}. Find all radially symmetric solutions of
(4.69) in R.
Solution: We look for solutions of
uxx + uyy = 0, in Ω, (4.81)
of the form
u(x, y) = f (r), for (x, y) ∈ R2 , (4.82)
where p
r= x2 + y 2 , (4.83)
2
and f : (0, ∞) → R is a C function.
Write the expression in (4.83) r2 = x2 + y 2 and differentiate on both sides
with respect to x, applying the Chain Rule to get
∂r
2r = 2x,
∂x
from which we get that
∂r x
= , for r > 0. (4.84)
∂x r
Similar calculations show that
∂r y
= , for r > 0, (4.85)
∂y r
Next, use the Chain Rule to obtain from (4.82) that
∂r
ux = f 0 (r) ,
∂x
so that, by virtue of (4.84),
x 0
ux = f (r), for r > 0. (4.86)
r
Similar calculations using (4.82) and (4.85) yield
y 0
uy = f (r), for r > 0. (4.87)
r
Next, use the Product Rule, the Quotient Rule, and the Chain Rule to obtain
from (4.86) that
∂r ∂r
1 0 rf 00 (r) − f 0 (r)
uxx = f (r) + x ∂x ∂x ;
r r2
thus, using (4.84),
1 0 x2 x2
uxx = f (r) + 2 f 00 (r) − 3 f 0 (r), for r > 0. (4.88)
r r r
Similar calculations, using (4.85) and (4.87) yield
1 0 y2 y2
uyy = f (r) + 2 f 00 (r) − 3 f 0 (r), for r > 0. (4.89)
r r r
Next, add the expressions in (4.88) and (4.89) to obtain
2 0 x2 + y 2 00 x2 + y 2 0
uxx + uyy = f (r) + f (r) − f (r), for r > 0,
r r2 r3
or using the fact that x2 + y 2 = r2 ,

2 0 1
uxx + uyy = f (r) + f 00 (r) − f 0 (r), for r > 0,
r r
or
1
uxx + uyy = f 00 (r) + f 0 (r), for r > 0. (4.90)
r
It follows from (4.90) that, if u in (4.82) solves Laplace’s equation in R, then f
solves the second order ODE
1
f 00 (r) + f 0 (r) = 0, for r > 0,
r
or
rf 00 (r) + f 0 (r) = 0, for r > 0,
which can be rewritten as
d
[rf 0 (r)] = 0, for r > 0. (4.91)
dr
Integrating the equation in (4.91) yields
rf 0 (r) = c1 , for r > 0,
and some constant c1 , or
c1
f 0 (r) = , for r > 0, (4.92)
r
and some constant c1 . Integrating the equation in (4.92) yields
f (r) = c1 ln r + c2 , for r > 0, (4.93)
and constants c1 and c2 .
It follows from (4.82) and (4.93) that radially symmetric solutions of (4.81)
are of the form
p
u(x, y) = c1 ln x2 + y 2 + c2 , for (x, y) 6= (0, 0), (4.94)
Example 4.2.2 (The Dirichlet Problem in an Annulus). For positive numbers,
r1 and r2 , with r1 < r2 , define Ω to be the annulus
p
Ω = {(x, y) ∈ R2 | r1 < x2 + y 2 < r2 }. (4.95)
Denote by Cr the circle of radius r centered at the origin.
Solve the boundary value problem:

 uxx + uyy = 0, in Ω;
u = a, on Cr1 ; (4.96)
u = b, on Cr2 ,

where a and b are real constants.

Solution: Since the annulus Ω in (4.95) has radial symmetry, and the
boundary conditions in (4.96) are also radially symmetric, it makes sense to
look for radially symmetric solutions of problem (4.96). According to the result
of Example 4.2.1, these are of the form given in (4.94); namely
p
u(x, y) = c1 ln x2 + y 2 + c2 , for (x, y) ∈ Ω, (4.97)
for some constants c1 and c2 .

The boundary conditions In (4.96) then imply that
c1 ln r1 + c2 = a (4.98)
and
c1 ln r2 + c2 = b, (4.99)
in view of (4.97). Solving the system of equations in (4.98) and (4.99) for c1
and c2 yields
b−a
c1 = ,
ln(r2 /r1 )
and
a ln r2 − b ln r1
c2 = .
ln(r2 /r1 )
Substituting these values for c1 and c2 into (4.97) yields a solution,
b−a p a ln r2 − b ln r1
u(x, y) = ln x2 + y 2 + , for (x, y) ∈ Ω, (4.100)
ln(r2 /r1 ) ln(r2 /r1 )
to the BVP in (4.96). The result of Problem 5 in Assignment #5 then shows

that the function u given in (4.100) is the solution of the BVP in (4.96).
4.2.2 Dilation Invariant Solutions to Laplace’s Equation

In this section we explore the effect of the change of variables

ξ α 0 x
= , (4.101)
η 0 β y
for nonzero real constants α and β, on the two–dimensional Laplace’s equation
uxx + uyy = 0, in R2 . (4.102)
The change of variables in (4.101) corresponds to

ξ = αx;
(4.103)
η = βy,
or
x = ξ/α;
(4.104)
y = η/β.
Setting
v(ξ, η) = u(x, y), (4.105)
where x and y are given in terms of ξ and η by the equations in (4.104), we
compute, using the Chain Rule, we obtain from (4.105) that
∂ξ ∂η
ux = vξ + vη ,
∂x ∂x
where, by virtue of the equations in (4.103),
∂ξ ∂η
=α and = 0,
∂x ∂x
so that
ux = αvξ . (4.106)
Similarly,
uy = βvη . (4.107)
Next, differentiate on both sides of (4.106) and apply the Chain Rules as in the
previous calculations to obtain
uxx = α2 vξξ . (4.108)
Similarly, we obtain from (4.107) that
uyy = β 2 vηη . (4.109)
Adding (4.108) and (4.109) we obtain
uxx uyy = α2 vξξ + β 2 vηη . (4.110)
Thus, if u solves Laplace’s equation in (4.102), we obtain from (4.110) that
α2 vξξ + β 2 vηη = 0. (4.111)
It follows from (4.111) that Laplace’s equation in R2 is invariant under the

scaling transformation in (4.101), provided that α2 = β 2 . We will therefore set
α = β = λ in (4.101) to get

ξ x
= Dλ , (4.112)
η y
where Dλ denotes the scalar matrix

λ 0
Dλ = ,
0 λ
for a nonzero parameter λ. The transformations in (4.112) form a one–parameter

family of dilations corresponding to the change of variables

ξ = λx;
(4.113)
η = λy.
Note from (4.113) that a combination of the variables that is independent of

the dilation parameter λ is
η y
= , for x 6= 0.
ξ x
This suggests that we look for solutions to Laplace’s equation in R2 of the form
y
u(x, y) = f , for x 6= 0, (4.114)
x
where f is a C 2 function of a single variable.
Set
y
s = , for x 6= 0, (4.115)
x
so that, in view of (4.114)
u(x, y) = f (s), (4.116)
where s is given by (4.115).
We look for a solution to Laplace’s equation in R2 of the form in (4.116)
where f is a C 2 function and s is as given in (4.115). Thus, assume that u
solves (4.102) and compute
∂s
ux = f 0 (s) , (4.117)
∂x
where we have used the Chain Rule and where
∂s y
= − 2, for x 6= 0,
∂x x
by virtue of (4.115), so that
∂s s
=− , for x 6= 0. (4.118)
∂x x
Substituting (4.118) into the right–hand side of (4.117) then yields
1
ux = − sf 0 (s), for x 6= 0. (4.119)
x
Next, differentiate with respect to x on both sides of (4.119) to get
1 0 1 ∂s 0 1 ∂s
uxx = sf (s) − f (s) − sf 00 (s) , for x 6= 0, (4.120)
x2 x ∂x x ∂x
where we have used the Product Rule and the Chain Rule. Then, substitute
(4.118) into the right–hand side of (4.120) to get
2s 0 s2
uxx = 2
f (s) + 2 f 00 (s), for x 6= 0. (4.121)
x x
Next, apply the Chain Rule to obtain from (4.116) that
∂s
uy = f 0 (s) , (4.122)
∂y
where
∂s 1
= , for x 6= 0. (4.123)
∂y x
It then follows from (4.122) and (4.123) that
1 0
uy = f (s), for x 6= 0. (4.124)
x
Differentiate on both sides of (4.124) with respect to y, apply the Chain Rule,
and use (4.123) to obtain
1 00
uyy = f (s), for x 6= 0. (4.125)
x2
Next, add the expressions in (4.121) and (4.125) to get
2s 0 1 + s2 00
uxx + uyy = 2
f (s) + f (s), for x 6= 0. (4.126)
x x2
It follows from (4.126) that, if u solves Laplace’s equation in R2 , then f solves
the second order ODE
2s 0 1 + s2 00
f (s) + f (s) = 0, for x 6= 0,
x2 x2
or
(1 + s2 )f 00 (s) + 2sf 0 (s) = 0. (4.127)
In order to solve the ODE in (4.127), set
v(s) = f 0 (s), (4.128)
so that
dv
(1 + s2 ) + 2sv = 0. (4.129)
ds
The first order ODE in (4.129) can be solved by separating variables to yield
Z Z
1 2s
dv = − ds,
v 1 + s2
or
1
ln |v| = ln + co , (4.130)
1 + s2
for some constant co .
Exponentiating on both sides of (4.130) and using the continuity of v we
obtain
c1
v(s) = , for s ∈ R, (4.131)
1 + s2
and some constant c1 . It follows from (4.128) and (4.131) that
c1
f 0 (s) = , for s ∈ R,
1 + s2
and some constant c1 , which can be integrated to yield
f (s) = c1 arctan(s) + c2 , for s ∈ R, (4.132)

It follows from (4.114) and (4.132) that dilation–invariant solutions of Laplace’s
equation in R2 are of the form
y
u(x, y) = c1 arctan + c2 , for x 6= 0, (4.133)
x
and constants c1 and c2 . The result in (4.133) states that dilation–invariant
harmonic functions in R2 are linear functions of the angle, θ, the the point
(x, y), for (x, y) 6= (0, 0), makes with the positive x–axis:
u = c1 θ + c2 ,
for constants c1 and c2 .
4.2.3 Dilation Invariant Solutions of the Diffusion Equa-

tion
In this section we look for dilation–invariant solutions of the one–dimensional
diffusion equation
∂u ∂2u
= D 2, for x ∈ R and t > 0, (4.134)
∂t ∂x
where D > 0 is the diffusivity constant. We proceed as in Section 4.2.2 by finding
conditions on parameters α and β so that the diffusion equation in (4.134) is
invariant under the change of variables

ξ = αx;
(4.135)
τ = βt,
where αβ 6= 0.
Write
v(ξ, τ ) = u(x, t), (4.136)
where x and t are given in terms of ξ and τ by inverting the system in (4.136),

x = ξ/α;
t = τ /β,
and use the Chain Rule to compute

∂ξ ∂τ
ux = vξ + vτ ,
∂x ∂x
∂ξ ∂η
=α and = 0,
∂x ∂x
so that
ux = αvξ . (4.137)
Similarly,
ut = βvτ . (4.138)
Next, differentiate on both sides of (4.137) and apply the Chain Rules as in the
previous calculations to obtain
uxx = α2 vξξ . (4.139)
Using the expressions in (4.138) and (4.139) we obtain
ut − D uxx = βvτ − Dα2 vξξ ,
so that, if u solves the diffusion equation in (4.134),
βvτ − Dα2 vξξ = 0. (4.140)
Hence, the diffusion equation in (4.134) is invariant under the change of variables
in (4.136) provided that
β = α2 . (4.141)
It follows from (4.140) and (4.141) that the diffusion equation in (4.134) is
invariant under the dilation

ξ = αx;
(4.142)
τ = α2 t.
It follows from (4.142) that combinations of the variables that are independent
of the dilation parameter, α, are
ξ2 x2 ξ x
= or √ =√ , for τ > 0 and t > 0.
τ t τ t
Thus, in order to find dilation–invariant solutions of the one–dimensional diffu-
sion equation, we look for solutions of the form

x
u(x, t) = f √ , for t > 0, (4.143)
t
where f is a C 2 function of a single variable.

Set
x
s= √ , for t > 0, (4.144)
t
so that, in view of (4.143)
u(x, t) = f (s), (4.145)
where s is given by (4.144).
Differentiate on both sides of (4.145) with respect to x, using the Chain

Rule, to get
∂s
ux = f 0 (s) ,
∂x
where
∂s 1
= √ , for t > 0, (4.146)
∂x t
by virtue of (4.144), so that
1
ux = √ f 0 (s), for t > 0. (4.147)
t
Differentiate with respect to x on both sides of (4.147), use the Chain Rule, and
the result in (4.146) to get
1 00
uxx = f (s), for t > 0. (4.148)
t
Next, differentiate on both sides of (4.145) with respect to t, using the Chain
Rule, to get
∂s
ut = f 0 (s) , (4.149)
∂t
where, by virtue of (4.144),
∂s x
= − √ ,
∂t 2t t
or, using (4.144),
∂s s
= − , for t > 0. (4.150)
∂t 2t
Substitute the result in (4.150) into the right–hand side of (4.149) to get
s 0
ut = − f (s), for t > 0. (4.151)
2t
It follows from (4.148) and (4.151) that, if u given in (4.145) solves the diffusion
equation in (4.134), then f solves the ODE
s 0 D
− f (s) = f 00 (s), for t > 0,
2t t
or
s 0
f 00 (s) +
f (s) = 0 (4.152)
2D
In order to solve the ODE in (4.152), set
v(s) = f 0 (s), (4.153)
so that
dv s
+ v = 0. (4.154)
ds 2D
The first order ODE in (4.154) can be solved by separating variables to yield
Z Z
1 s
dv = − ds,
v 2D
or
s2
ln |v| = − + co , (4.155)
4D
for some constant co .
Exponentiating on both sides of (4.155) and using the continuity of v we
obtain 2
v(s) = c1 e−s /4D , for s ∈ R, (4.156)
and some constant c1 . It follows from (4.156) and (4.153) that
2
f 0 (s) = c1 e−s /4D
, for s ∈ R,
and some constant c1 , which can be integrated to yield

Z s
2
f (s) = c1 e−z /4D dz + c2 , for s ∈ R, (4.157)
0
and constants c1 and c2 . It follows from (4.143) and (4.157) that dilation–
invariant solutions of one–dimensional diffusion equation in (4.134) are of the
form
Z x/√t
2
u(x, t) = c1 e−z /4D dz + c2 , for x ∈ R and t > 0, (4.158)
0

Chapter 5
Solving Linear PDEs
In Chapter 4 we saw two general approaches for finding solutions to first or

second order PDEs: using characteristic curves and looking for symmetric solu-
tions. In theory, these methods could be applied to nonlinear or linear equations.
In this chapter we explore methods that exploit the special structure provided
by linear PEDs. In Section 3.1 we saw the Principle of Superposition (Proposi-
tion 3.1.1 on page 46 in these notes), which states that linear combinations of
solutions to the homogeneous linear PDE
Lu = 0,
where L is a linear differential operator, are also solutions. Thus, in principle,

we can use superposition to construct solutions of linear PDEs satisfying certain
conditions by putting together known solutions. We will see in this chapter that
this procedure can be carried out by adding together infinitely many solutions
in the form of a series or and integral transform. We will begin by solving the
vibrating string equation that we derived in in Section 2.3.3 and Section 2.4
using separation of variables and Fourier series expansions. In this chapter, we
will also look at a a special solutions that can be used as building blocks to
obtain solutions to initial and/or boundary value problems for a large class of
linear PDEs. These special solutions are known as Fundamental Solutions.
5.1 Solving the Vibrating String Equation

In this section we construct a solution of the initial–boundary value problem for
the one–dimensional, linear, homogeneous wave equation in (2.143); namely,
utt − c2 uxx = 0, for x ∈ (0, L) and t > 0, (5.1)
where c is a positive constant, subject to the boundary conditions
u(0, t) = 0, for t > 0, (5.2)
77
78 CHAPTER 5. SOLVING LINEAR PDES
and
u(L, t) = 0, for t > 0, (5.3)
and the initial conditions
u(x, 0) = f (x), for x ∈ [0, L], (5.4)
and
ut (x, 0) = 0, for x ∈ [0, L], (5.5)
where f : [0, L] → R is a given, real valued, continuous function defined on the
interval [0, L].
We recall that the constant c in (5.1) is given by
τ
c2 = , (5.6)
ρ
where ρ is the constant density of the string in units of mass per length, and τ
is the constant tension of the string in units of force.
The PDE in (5.1), together with the boundary conditions in (5.2) and (5.3),
and the initial conditions in (5.4) and (5.5), models the small amplitude vibra-
tions of a taut string of length L that is fixed at the end–points 0 and L of the
interval [0, L], that is plucked from rest from an initial shape given by the graph
of the function f . In this section we construct a solution of this initial–boundary
value problem, which we write in concise form as


 utt − c2 uxx = 0, for x ∈ (0, L) and t > 0;

u(0, t) = u(L, t) = 0, for t > 0;
(5.7)


 u(x, 0) = f (x), for x ∈ [0, L];
ut (x, 0) = 0, for x ∈ [0, L].

At this point, we will assume that f is a continuous function that satisfies
f (0) = 0 and f (L) = 0.
As we refine the construction of a solution u : [0, L] × [0, ∞) → R of the system

in (5.7), we will make further assumptions on f .
We look for solutions, u : [0, L] × [0, ∞) → R, of the partial differential
equation in (5.1) that have continuous, second partial derivatives with respect
to t and with respect to x. We first consider functions of a special form
u(x, t) = y(x)h(t), for x ∈ [0, L], and t > 0, (5.8)
where y ∈ C 2 ([0, L], R) and h ∈ C 2 ([0, ∞), R); that is y : [0, L] → R is a

real valued function defined on [0, L] that has continuous second derivative
in (0, L), which can be extended to a continuous function on [0, L]; similarly,
h : [0, ∞) → R is a real valued, continuous function defined on [0, ∞) that is
twice differentiable in (0, ∞) with continuous second derivative that can be
5.1. SOLVING THE VIBRATING STRING EQUATION 79
extended to a continuous function on [0, ∞). We will also require that these
solutions satisfy the boundary conditions in (5.2) and (5.3).
Once we find solutions of (5.1) of the form given in (5.8) that satisfy the
boundary conditions in (5.2) and (5.3), we take advantage of the fact that the
PDE in (5.1) is linear to construct linear combinations of these solutions, which
will also be solutions, to approximate solutions that satisfy the initial conditions
in (5.4) and (5.5). This is the general outline of the procedure to be used in
this section, which is know in the literature as the method of separation of
variables.
5.1.1 Separation of Variables

Suppose that u : [0, L]×[0, ∞) → R is of the form in (5.8), where y ∈ C 2 ([0, L], R)
and h ∈ C 2 ([0, ∞), R). Then, the second partial derivatives, utt and uxx , of u
are given by
uxx (x, t) = y 00 (x)h(t), for x ∈ (0, L), and t > 0, (5.9)
and
utt (x, t) = y(x)h00 (t), for x ∈ (0, L), and t > 0. (5.10)
If we require that u be a solution of the PDE in (5.1), substituting the expression
for uxx in (5.9) and for utt in (5.10) into (5.1) then yields
y(x)h00 (t) = c2 y 00 (x)h(t), for x ∈ (0, L), and t > 0. (5.11)
Next, divide on both sides of the equation in (5.11) by c2 y(x)h(t) (assuming

that c2 y(x)h(t) 6= 0) to obtain
h00 (t) y 00 (x)

= , for x ∈ (0, L), and t > 0, (5.12)
c2 h(t) y(x)
provided that y(x) 6= 0 and h(t) 6= 0.

For the expression in (5.12) to be true for all x ∈ (0, L) and t > 0 at which
the denominators in (5.12) are not 0, it must be the case that both sides of the
equation in (5.12) must be equal to a constant. The reason for this is that the
variables x and t vary independently of each other. Denoting that constant by
−λ, we obtain that
y 00 (x)
= −λ,
y(x)
and
h00 (t)
= −λ,
c2 h(t)
from which we obtain that pair of ordinary differential equations
y 00 (x) + λy(x) = 0, for x ∈ (0, L), (5.13)

and
h00 (t) + λc2 h(t) = 0, for t > 0. (5.14)
Next, if u is to satisfy the boundary conditions in (5.2) and (5.3), it must
be the case that
y(0)h(t) = 0 and y(L)h(t) = 0, for all t > 0. (5.15)
At this point we make the observation that the function defined by
u(x, t) = 0, for all x ∈ [0, L] and t > 0, (5.16)
is a solution of the PDE in (5.1). We call it the trivial solution of the equation.
We also note that the function in (5.16) satisfies the boundary conditions in (5.2)
and (5.3) and the initial condition in (5.5). However, for a general function
f : [0, L] → R, the trivial function in (5.16) does not satisfy the initial condition
in (5.4). Thus, if we are to construct a solution of the initial–boundary problem
in (5.7), it cannot be the trivial solution. Hence, we look for nontrivial solutions
of the PDE in (5.1) of the form in (5.8).
Consequently, if u ∈ C 2 ([0, L] × [0, ∞), R) of the form given in (5.8) is a
nontrivial solution of the PDE in (5.1) that also satisfies the boundary conditions
in (5.15), then it must be the case that
y(0) = 0 and y(L) = 0. (5.17)
To see why this is the case, use the assumption that u is nontrivial to conclude
that there exists to > 0 such that h(to ) 6= 0 and, according to (5.15),
y(0)h(to ) = 0 and y(L)h(to ) = 0. (5.18)
We can see now that (5.17) follows from (5.18) and the fact that h(to ) 6= 0.
We therefore look for nontrivial solutions of the the two–point, boundary
value problem (BVP)
 00
 y + λy = 0, for x ∈ (0, L);
y(0) = 0; (5.19)
y(L) = 0.

The BVP in (5.19) is an example of an eigenvalue problem. We need to

determine values of λ for which the BVP has nontrivial solutions. These values
are called eigenvalues. The corresponding nontrivial solutions will be called
eigenfunctions.
We consider three possibilities for λ in the BVP in (5.19): λ = 0, λ < 0, or
λ > 0.
If λ = 0, the ODE in (5.19) becomes
y 00 = 0,
which has general solution
y(x) = ax + b, for x ∈ R, (5.20)
and for some constants a and b.

Since the function y given in (5.20) also needs to satisfy the boundary con-
dition
y(0) = 0
in the BVP (5.19), it follows from (5.20) that
b = 0,
y(x) = ax, for x ∈ R. (5.21)
By the same token, since y also has to satisfy the boundary condition
y(L) = 0
in (5.19), we get from (5.21) that
aL = 0,
we get that
a = 0,
since L > 0. Consequently, in view of (5.21),
y(x) = 0, for x ∈ R,
which shows that y must be the trivial solution. Consequently, if λ = 0, the

BVP in (5.19) has only the trivial solution. Therefore, λ = 0 is not an eigenvalue
of the BVP (5.19).
Next, assume that λ < 0 in BVP (5.19), and set λ = −µ2 , where µ > 0.
Then, the ODE in (5.19) becomes
y 00 − µ2 y = 0. (5.22)
The characteristic equation of the ODE in (5.22) (see Appendix A.1) is
m2 − µ2 = 0,
which has roots µ and −µ; hence, according to (A.5) in Appendix A.1, the
general solution of the ODE in (5.22) is
y(x) = c1 eµx + c2 e−µx , for x ∈ R, (5.23)
for constants c1 and c2 .

The boundary conditions in the BVP in (5.19) imply from (5.23) that

c1 + c2 = 0;
c1 eµL + c2 e−µL = 0,
which we can write in matrix form as

1 1 c1 0
= . (5.24)
eµL e−µL c2 0
The homogeneous system in (5.24) has only the trivial solution if and only if

1 1
det µL 6= 0, (5.25)
e e−µL
where
1 1
det = e−µL − eµL .
eµL e−µL

1 1
det =0
eµL e−µL
if and only if
eµL = e−µL
if and only if
e2µL = 1
if and only if
2µL = 0,
from which we conclude that µ = 0, which contradicts the assumption that
µ > 0. Hence, (5.25) must hold true, from which we conclude that the system
in (5.24) has only the trivial solution
c1 = c2 = 0;
y(x) = 0, for all x ∈ R,
which shows that, if λ < 0, then the BVP in (5.19) has only the trivial solution.
It remains to consider the case λ > 0. Thus, assume that λ = ω 2 , where
ω > 0. In this case, the ODE in BVP (5.19) becomes
y 00 + ω 2 y = 0, (5.26)
which has characteristic equation
m2 + ω 2 = 0,
with complex conjugate roots ωi and −ωi. Consequently, according to (A.5) in

Appendix A.1, the general solution of the ODE in (5.26) is given by
y(x) = c1 cos ωx + c2 sin ωx, for all x ∈ R. (5.27)

The boundary condition y(0) = 0 in BVP (5.19) implies from (5.27) that c1 = 0;
y(x) = c2 sin ωx, for all x ∈ R. (5.28)
The second boundary condition in (5.19) implies from (5.28) that
c2 sin ωL = 0. (5.29)
It follows from (5.28) that c2 6= 0; otherwise y would be trivial. Hence, we

obtain from (5.29) that
sin ωL = 0. (5.30)
The equation in (5.30) has infinitely many solutions given by
ωL = nπ, for n ∈ Z,

nπ
ω= , for n ∈ Z. (5.31)
L
Since we are assuming that ω > 0, we have to exclude the negative solutions
and 0 from the list in (5.31); so that,
nπ
ω= , for n = 1, 2, 3, . . . (5.32)
L
Next, since λ = ω 2 , we have shown that the eigenvalue problem in (5.19) has
an infinite sequence, (λn ), of positive eigenvalues given by
nπ 2
λn = , for n = 1, 2, 3, . . . (5.33)
L
To each of the eigenvalues, λn , in (5.33) there correspond eigenfunctions, yn : [0, L] →
R, defined by
nπx
yn (x) = cn sin , for x ∈ [0, L] and n = 1, 2, 3, . . . , (5.34)
L
according to (5.28), for arbitrary non–zero constants cn .
Next, we turn to the ODE in (5.14) where λ is replaced by λn for n ∈ N.
We therefore get a sequence of ODEs
h00 (t) + λn c2 h(t) = 0, for t > 0 and n = 1, 2, 3, . . . (5.35)
The characteristic equation for each of the ODEs in (5.35) is
m2 + λn c2 , for n = 1, 2, 3, . . . ,
√ √
which has complex conjugate roots λn ci and − λn ci. Consequently, by (A.6)
in Appendix A, the general solutions of the ODEs in (5.35) are given by

nπct nπct
hn (t) = an cos + bn sin , (5.36)
L L
for t > 0, n = 1, 2, 3, . . ., and arbitrary constants an and bn , for n ∈ N, where

we have used (5.33).
Thus, in view of (5.8), (5.34) and (5.36), for each n ∈ N, there is a solution
of the wave equation in (5.1) of the form
un (x, t) = yn (x)hn (t), for x ∈ [0, L] and t > 0,
or, takin cn = 1 for all n ∈ N in (5.34),
nπx
nπct

nπct

un (x, t) = sin an cos + bn sin , (5.37)
L L L
for x ∈ [0, L], t > 0, n = 1, 2, 3, . . ., and arbitrary constants an and bn , for
n ∈ N, which also satisfies the boundary conditions in the initial–boundary
value problem in (5.7).
Now, each of the functions in (5.37) does not, in general, satisfy the initial
conditions in (5.7); unless, for exmaple, the function f in (5.7) is a of a very
spacial form: πx
f (x) = a1 sin , for x ∈ [0, L],
L
where |a1 | is very small, u1 (x, t) in (5.37) with b1 = 0, no function in (5.37) by
itself will yield a solution of the initial–boundary value problem (5.7). However,
since the PDE in (5.7) is linear, sums of the functions in (5.37) are also solutions
of the PDE. We therefore consider next functions of the form
N nπx
X nπct nπct
uN (x, t) = sin an cos + bn sin , (5.38)
n=1
L L L
for x ∈ [0, L], t > 0, N = 1, 2, 3, . . ., and arbitrary constants an and bn , for

n ∈ N. We note that each of the functions in (5.38) satisfies the boundary
conditions in problem (5.7), since
uN (0, t) = 0 and uN (L, t) = 0, for all t > 0.
Next, we explore whether functions in (5.38) can satisfy the initial conditions
in (5.7) as well.
From (5.38) we can compute the partial derivative
N nπx
∂ X nπc nπct nπct
[uN (x, t)] = sin −an sin + bn cos ,
∂t n=1
L L L L
(5.39)
for x ∈ [0, L] and t > 0. Thus, if we set bn = 0 for all n, we get from (5.39) that
∂
[u (x, 0)] = 0, for all x ∈ [0, L]. (5.40)
∂t N
It follows from (5.40) that the functions
N nπx
X nπct
uN (x, t) = an sin cos , for x ∈ [0, L], t > 0, (5.41)
n=1
L L
N = 1, 2, 3, . . ., and arbitrary constants an , satisfy the second of the initial

conditions in the problem (5.7), as well as the boundary conditions and the
PDE in that problem. However, in general, the functions in (5.41) do not
satisfy first initial condition in (5.7), unless
N
X nπx
an sin = f (x), for x ∈ [0, L], (5.42)
n=1
L
for some N ∈ N and some constants an . Only if the function f is a linear

combination of the trigonometric functions on the left–hand side of (5.42), will
the function given in (5.41) be a solution of the initial–boundary value problem
for the one–dimensional wave equation in (5.7). For a large class of functions
f ∈ C([0, L], R), though, we will be able to express the f as a limit linear
combinations of trigonometric functions as in the left–hand side of (5.42).
Set
XN nπx
SN (x) = cn sin , for x ∈ [0, L], (5.43)
n=1
L
where the coefficients cn will be chosen in a special way, depending on f , to be
described in the next section. We would like to know when
lim SN (x) = f (x), for x ∈ [0, L]. (5.44)

N →∞
If the limit on the left–hand side of the (5.44) exists, we denote it by

∞
X nπx
cn sin , for x ∈ [0, L], (5.45)
n=1
L
and call it a Fourier series, or a Fourier series expansion. Thus, in order

to construct a solution of the initial–boundary value problem in (5.7), we first
need to answer the question of whether or not the function f ∈ C([0, L], R) has
a Fourier series expansion; that is, the limit expression in (5.44), where SN is
given in (5.43), holds true (in some appropriate way), or
∞
X nπx
cn sin = f (x), for x ∈ [0, L], (5.46)
n=1
L
for a special class of coefficients, cn , determined by f . We will answer this

question in the following section.
5.1.2 Fourier Series Expansions

Let’s assume for a moment that the series on the left–hand side of (5.46) con-
verges uniformly to to f on [0, L], so that term–by–term integration of the series
is justified. For each m ∈ N, multiply both sides of the equation in (5.46) by
mπx
sin
L
for some m ∈ N to get

∞
X nπx mπx mπx
cn sin sin = f (x) sin , for x ∈ [0, L],
n=1
L L L
and integrate on both sides from 0 to L to get

Z LX ∞ nπx mπx Z L mπx
cn sin sin dx = f (x) sin dx. (5.47)
0 n=1 L L 0 L
The assumption that the series in the integrand on the left–hand side of (5.47)
justifies the term–by–term integration of the series to get
X∞ Z L nπx mπx Z L mπx
cn sin sin dx = f (x) sin dx. (5.48)
n=1 0 L L 0 L
To evaluate the integrals on the left–hand side on (5.48), we use the trigono-
metric identity
1
sin A sin B = [cos(A − B) − cos(A + B)] , for A, B ∈ R. (5.49)
2
Applying (5.49) to the integrands on the left–hand side of (5.48), we get that
nπx mπx 1
(n − m)πx

(n + m)πx

sin sin = cos − cos , (5.50)
L L 2 L L
for m, n ∈ N. We integrate each of the terms on the right–hand side of (5.50)
from 0 to L separately. First, integrate the right–most term in (5.50) to get
Z L L
(n + m)πx L (n + m)πx
cos dx = sin = 0,
0 L (n + m)π L 0
for all n, m ∈ N; so that, in view of (5.50),

Z L
1 L

(n − m)πx
nπx mπx Z
sin sin dx = cos dx, (5.51)
0 L L 2 0 L
for n, m ∈ N.
To evaluate the integral on the right–hand side of (5.51), we consider the
cases n 6= m and n = m separately.
Suppose that n 6= m and evaluate
Z L L
(n − m)πx L (n + m)πx
cos dx = sin = 0;
0 L (n − m)π L 0

Z L nπx mπx
sin sin dx = 0, if n 6= m. (5.52)
0 L L
On the other hand, if n = m, it follows from (5.51) that

Z L nπx mπx L
sin sin dx = , if n = m. (5.53)
0 L L 2
Combining (5.52) and (5.53) we then have that
L

Z L nπx mπx  ,
 if n = m;
 2
sin sin dx = (5.54)
0 L L 

0, if n 6= m.

Using the result in (5.54), we obtain from (5.48) that

Z L
L mπx
cm = f (x) sin dx,
2 0 L
from which we get the formula
2 L
Z nπx
cn = f (x) sin dx, for n ∈ N, (5.55)
L 0 L
for computing the coefficients in the Fourier series expansion for f in (5.46).
Given f ∈ C([0, L], R), define
N
X nπx
SN (x) = cn sin , for x ∈ [0, L], (5.56)
n=1
L
where the coefficients cn , for n ∈ N, are given in (5.55).

In the remainder of this section, we determine conditions on f ∈ C([0, L], R)
for which
lim SN (x) = f (x), for x ∈ [0, L], (5.57)
N →∞
where SN is given in (5.56), and for which the convergence in (5.57) is uniform,
so that the calculations leading to (5.55) are justified.
If the statement in (5.57) is true, we write
∞
X nπx
cn sin = f (x), for x ∈ [0, L], (5.58)
n=1
L
where the coefficients cn , for n ∈ N, are given in (5.55).

The expression on the left–hand side of (5.58) is an example of a Fourier
series expansion.
In general, assume that f : R → R is a bounded, periodic function of period
2L, where L > 0. Assume also that f is integrable over [−L, L]; so that,
Z L
|f (x)| dx < ∞, (5.59)
−L
where the integral in (5.59) denotes the Riemann integral. If this is the case,
we can the define Fourier coefficients of f as follows:
Z L
1
ao = f (x) dx; (5.60)
2L −L
Z L
1 nπx
an = f (x) cos dx, for n = 1, 2, 3, . . . ; (5.61)
L −L L
and
Z L
1 nπx
bn = f (x) sin dx, for n = 1, 2, 3, . . . . (5.62)
L −L L
The Fourier series expansion of f is the trigonometric series

∞ h
X nπx nπx i
ao + an cos + bn sin , for x ∈ [−L, L], (5.63)
n=1
L L
where the coefficients an , for n = 0, 1, 2, 3, . . ., and bn , for n = 1, 2, 3, . . ., are

defined in (5.60), (5.61) and (5.62).
Example 5.1.1. Let f : [0, L] → R be the function giving the first initial con-
dition in the problem (5.7). Since we are assuming that f (0) = f (L) = 0, f can
be extended to a continuous, odd, periodic function of period 2L be defining
f (x) = −f (−x), for − L 6 x < 0,
and
f (x + 2L) = f (x), for all x ∈ R.
In this case, the formula for the Fourier coefficients of f in (5.60), (5.61) and
(5.62 yield
an = 0, for n = 0, 1, 2, 3, . . .
and
Z L
2 nπx
bn = f (x) sin dx, for n = 1, 2, 3, . . . , (5.64)
L 0 L
since f is odd and therefore the integrand in (5.62) is even.
Thus, in this case, the Fourier series expansion of f yields
∞
X nπx
bn sin , for x ∈ [0, L],
n=1
L
where the coefficients bn , for n ∈ N are given in (5.64). This is the same
expansion given in (5.46) with cn in place of bn .
For each N = 1, 2, 3, . . ., put

N h
X nπx nπx i
SN (x) = ao + an cos + bn sin , for x ∈ [−L, L], (5.65)
n=1
L L
where an , for n = 0, 1, 2, 3, . . ., and bn , for n = 1, 2, 3, . . ., are the Fourier

coefficients of f in (5.60), (5.61) and (5.62.
We would like to know which conditions on the function f will guarantee
that
lim SN (x) = f (x), for x ∈ [0, L], (5.66)
N →∞
where SN is given by (5.65). If (5.66) holds true, we say that the Fourier series
expansion for f given in (5.63) converges to f pointwisely, and write
X∞ h nπx nπx i
ao + an cos + bn sin = f (x), for x ∈ [−L, L]. (5.67)
n=1
L L
We would also like to determine conditions on f under which the convergence

in (5.66) is uniform.
Substitute the expressions for an , for n = 0, 1, 2, 3, . . ., and bn , for n ∈ N,
given in (5.60)–(5.62 into the expression for SN in (5.65) to get
Z L
1
SN (x) = f (y) dy
2L −L
N
!
Z L
X 1 nπy nπx
+ f (y) cos dy cos (5.68)
n=1
L −L L L
N
!
Z L
X 1 nπy nπx
+ f (y) sin dy sin ,
n=1
L −L L L
for x ∈ [0, L], where we have used y as the variable of integration in the definition
of the Fourier coefficients because the variable x is being used as the argument
of SN .
Next, interchange summation and integration and use the distributive prop-
erty to in the last two terms on the right–hand side of (5.68) to get
Z L N h
1 X nπx nπy nπx nπy i
f (y) cos cos + sin sin dy. (5.69)
L −L n=1
L L L L
The term in the summation in (5.69) can be simplifies using the trigonometric
identity
cos(A − B) = cos A cos B + sin A sin B, for A, B ∈ R

to yield
L N
nπ(x − y)
Z
1 X
f (y) cos dy. (5.70)
L −L n=1
L
Consequently, the expression in (5.68) can be rewritten as
N !
1 L

nπ(x − y)
Z
1 X
SN (x) = f (y) + cos dy, for x ∈ [0, L], (5.71)
L −L 2 n=1 L
which can be rewritten in terms of the function

N
1 X nπθ
DN (θ) = + cos , for θ ∈ R, (5.72)
2 n=1 L
as Z L
1
SN (x) = f (y)DN (x − y) dy, for x ∈ [0, L]. (5.73)
L −L
The function defined in (5.72) is called the Dirichlet kernel. Observe that
DN is even and periodic of period 2L. Furthermore, it can be shown that

1 πθ
sin N +
2 L
DN (θ) = , for θ 6= 0. (5.74)
πθ
2 sin
2L
(See Appendix C.1 for a derivation of the result in (5.74)).

Note that, according to (5.72),
1
DN (0) = N + . (5.75)
2
Using (5.72) we can compute
Z L
DN (θ) dθ = L, for all N ∈ N. (5.76)
−L
Next, make the change of variables
z =x−y
in the integral on the right–hand side of (5.73) to get

Z x−L
1
SN (x) = f (x − z)DN (z) (−dz), for x ∈ [−L, L],
L x+L
or Z x+L
1
SN (x) = f (x − z)DN (z) dz, for x ∈ [−L, L];
L x−L
so that,
Z L
1
SN (x) = f (x − z)DN (z) dz, for x ∈ [−L, L], (5.77)
L −L
where we have used the 2L–periodicity of the integrand.

We will first prove, using the representation in (5.77), that for the case
in which f is assumed to be a C 1 function, (5.66) holds true; that is, the
trigonometric expansion of f in (5.65), where the Fourier coefficients an , for
n = 0, 1, 2, 3, . . ., and bn , for n ∈ N, given in (5.60)–(5.62, converges to f
pointwisely in [−L, L]. We will state this fact as a theorem.
Theorem 5.1.2 (Point–wise Convergence.). Let f ∈ C 1 (R, R) be a bounded,
periodic function of period 2L, and let
N h
X nπx nπx i
SN (x) = ao + an cos + bn sin , (5.78)
n=1
L L
for x ∈ [−L, L] and N = 1, 2, 3, . . ., where an , for n = 0, 1, 2, 3, . . ., and bn , for

n ∈ N, are given in (5.60)–(5.62). Then,
lim SN (x) = f (x), for x ∈ [−L, L]. (5.79)

N →∞
In the proof of Theorem 5.1.2 we will use the following result known as the
Riemann–Lebesgue Lemma. We state here a spacial form of the result in the
context of 2L–periodic functions and the Fourier coefficients.
Theorem 5.1.3. Assume that F : R → R is bounded, 2L–periodic, and inte-
grable over [−L, L]. Let an , for n = 0, 1, 2, 3, . . ., and bn , for n = 1, 2, 3, . . .,
denote the Fourier coefficients of F given in (5.60)–(5.62), with F in place of F .
Then,
lim an = 0 and lim bn = 0. (5.80)
n→∞ n→∞
We restate the conclusion of the Riemann–Lebesgue Lemma in the form that

will be applied in the proof of Theorem 5.1.2.
Z L
N πy
lim F (y) cos dy = 0. (5.81)
N →∞ −L L
and Z L
N πy
lim F (y) sin dy = 0. (5.82)
N →∞ −L L
A more general version of the Riemann–Lebesgue Lemma is the following.
Theorem 5.1.4. Assume that F : [a, b] → R is absolutely integrable on [a, b];
so that,
Z b
|F (x)| dx < ∞.
a
Then,
Z b
lim F (x) cos(λx) dx = 0, (5.83)
|λ|→∞ a
and Z b
lim F (x) sin(λx) dx = 0. (5.84)
|λ|→∞ a
For a proof of this result, see [Tol62]. We presents two examples that the
the result in Theorem 5.1.4 plausible.
Example 5.1.5. Let F (x) = 1 for all x ∈ R and compute

Z b Z b
F (x) cos(λx) dx = cos(λx) dx
a a
b
1
= sin(λx) ;
λ a
so that
Z b
1
F (x) sin(λx) dx = [sin(λb) − sin(λa)], for λ 6= 0. (5.85)
a λ
Take absolute value of both sides of (5.85) and apply the triangle inequality to
get
Z b
2
F (x) cos(λx) dx 6 , for λ 6= 0,
a |λ|
from which (5.83) follows for the case F (x) = 1 for all x.
Similar calculations show that (5.84) is also true in this case.
We present a second example as a proposition that will used later in this

section when we prove the uniform convergence of Fourier series of C 1 functions.
Proposition 5.1.6. Assume that F : [a, b] → R is bounded on [a, b] and differ-

entiable on (a, b) and that
Z b
|F 0 (x)| dx < ∞; (5.86)
a
that is, F 0 is absolutely integrable over [a, b]. Then, (5.83) and (5.84) hold true.
Proof: Use integration by parts to compute

Z b b Z b
1 1
F (x) cos(λx) dx = F (x) sin(λx) − F 0 (x) sin(λx) dx,
a λ a λ a
or Z b
1
F (x) cos(λx) dx = [F (b) sin(λb) − F (a) sin(λa)]
a λ
(5.87)
Z b
1 0
− F (x) sin(λx) dx.
λ a
Next, take absolute value of both sides of (5.87) and apply the triangle inequality
to get
Z b Z b
2M 1
F (x) cos(λx) dx 6 + |F 0 (x)| dx, for λ 6= 0, (5.88)
a |λ| |λ| a
where M is an upper bound of F on [a, b]. In view on (5.86), we can see that
(5.83) follows from (5.88).
Similar calculations show that (5.84) is also true in this case.
Proof of Theorem 5.1.2: Using (5.76) we see that
1 L
Z
f (x) = f (x)DN (z) dz, for x ∈ [−L, L]. (5.89)
L −L
Then, using the representation for SN in (5.77) together with (5.89),
1 L
Z
SN (x) − f (x) = [f (x − z) − f (x)]DN (z) dz, for x ∈ [−L, L], (5.90)
L −L
Next, we use the formula for the Dirichlet kernel in (5.74) to write (5.90) as
1 L f (x − z) − f (x)
Z
1 πz
SN (x) − f (x) = sin N + dz, (5.91)
L −L 2 sin πz

2 L
2L
for x ∈ [−L, L].
Write
f (x − z) − f (x)
G(x, z) = πz , for z 6= 0 and x, z ∈ [−L, L], (5.92)
2 sin
2L
and observe that, using L’Hospital’s Rule,
f (x − z) − f (x) L
lim πz = − f 0 (x),
z→0
2 sin π
2L
where we have used the assumption that f is a C 1 function. Consequently,
defining
L
G(x, 0) = − f 0 (x), for x ∈ [−L, L], (5.93)
π
We then see that the function G defined in (5.92) and (5.93), for x ∈ [−L, L]
and z ∈ [−L, L], is continuous on [−L, L] × [−L, L]. (Observe that denominator
in (5.92) is zero only at z = 0 in [−L, L].) Thus, the function G is bounded and
continuous on [−L, L] × [−L, L], and therefore the map
z 7→ G(x, z), for z ∈ [−L, L]
is absolutely integrable for each x ∈ [−L, L]. (We are aiming here at applying
the Riemann–Lebesgue Lemma as stated in (5.81) and (5.82).)
Using G(x, z) as defined in (5.92) and (5.93), we can rewrite (5.91) as
Z L
1 1 πz
SN (x) − f (x) = G(x, z) sin N+ dz,
L −L 2 L
for x ∈ [−L, L], which we can in turn write as

Z L
1 πz N πz
SN (x) − f (x) = G(x, z) cos sin dz
L −L 2L L
(5.94)
Z L
1 πz N πz
+ G(x, z) sin cos dz,
L −L 2L L
for x ∈ [−L, L].

Next, observe that, for each x ∈ [−L, L],
Z L πz Z L
G(x, z) cos dz 6 |G(x, z)| dz < ∞,
−L 2L −L
since G(x, z) is absolutely integrable over [−L, L] for each x ∈ [−L, L]. Thus,
we can apply the result of the Riemann–Lebesgue Lemma in (5.82) to deduce
that
Z L πz
N πz
lim G(x, z) cos sin dz = 0, for each x ∈ [−L, L]. (5.95)
N →∞ −L 2L L
Similar calculations can be used to show that

Z L πz
N πz
lim G(x, z) sin cos dz = 0, for each x ∈ [−L, L]. (5.96)
N →∞ −L 2L L
Combining (5.94), (5.95) and (5.96 we conclude that
lim (SN (x) − f (x)) = 0, for all x ∈ [−L, L],

N →∞
from which (5.79) follows. We have therefore completed the proof of Theorem
5.1.2.
We will next show that then the convergence of SN to f in (5.79) is actually

uniform convergence over [−L, L]. This means that, for any ε > 0, there exists
Nε ∈ N, which depends only on ε, such that
N > Nε ⇒ |SN (x) − f (x)| < ε, for all x ∈ [−L, L]. (5.97)
We will prove this as a consequence of the fact that f ∈ C 1 (R, R); so that, f 0 is
bounded on [−L, L]; that is, there exists a constant M > 0 such that
|f 0 (x)| 6 M, for all x ∈ [−L, L]. (5.98)
Theorem 5.1.7 (Uniform Convergence 1). Let f ∈ C 1 (R, R) be a bounded,

periodic function of period 2L. Then, the sequence of functions (SN ) given in
(5.78) converges uniformly to f over [−L, L].
Proof: As in the proof of Theorem 5.1.2, we begin with

Z L
1
SN (x) − f (x) = [f (x − z) − f (x)]DN (z) dz, for x ∈ [−L, L]. (5.99)
L −L
Our goal is to show that, given any ε > 0, we can find Nε ∈ N such that the
absolute value of the integral on the right–hand side of (5.99) is less than εL for
all N > Nε and all x ∈ [−L, L].
Before we proceed any further, let’s make the change of variables y = −z in
the integral on the right–hand side of (5.99) and use the fact that the Dirichlet
kernel DN is an even function to write
Z L
1
SN (x) − f (x) = [f (x + y) − f (x)]DN (y) dy, for x ∈ [−L, L]. (5.100)
L −L
Let ε > 0 be given and let δ > 0 (to be chosen later) be such that δ < L.
Write the expression on the right–hand side of (5.100) as the sum of the
three terms
1 −δ
Z
I1 (x) = [f (x + y) − f (x)]DN (y) dy, for x ∈ [−L, L], (5.101)
L −L
Z δ
1
I2 (x) = [f (x + y) − f (x)]DN (y) dy, for x ∈ [−L, L], (5.102)
L −δ
and
Z L
1
I3 (x) = [f (x + y) − f (x)]DN (y) dy, for x ∈ [−L, L]; (5.103)
L δ
so that,
SN (x) − f (x) = I1 (x) + I2 (x) + I3 (x), for x ∈ [−L, L]. (5.104)

We estimate each of the quantities I1 , I2 and I3 in (5.101), (5.102) and

(5.103), respectively, separately. We begin with I2 in (5.102).
Use integration by parts to compute
Z δ
δ
[f (x + y) − f (x)]DN (y) dy = [[f (x + y) − f (x)]VN (y)]−δ
−δ
(5.105)
Z δ
− f 0 (x + y)VN (y) dy,
−δ
for x ∈ [−L, L], where

Z y
VN (y) = DN (z) dz, for y ∈ [−L, L]. (5.106)
0
We derive a few properties of the function VN defined in (5.106). In particular,

we show that VN is bounded independently of N ; that is, there exists a constant
M1 > 0 such that
|VN (y)| 6 M1 , for all y ∈ [−L, L] and all N ∈ N. (5.107)
We first note that, since DN is an even function, then VN is odd.

Compute, using (5.74),

1 πz
Z y sin N +
2 L
VN (y) = πz dz
0 2 sin
2L

1
sin N + r
L πy/L
Z
2
= dr,
π 0 2 sin(r/2)
πz
where we have made the change of variables: r = . We therefore have that
L
L πy/L
Z
1 1
VN (y) = sin N + r dr,
π 0 2 2 sin(r/2)
L πy/L
Z
1 1 1
VN (y) = sin N + r − dr
π 0 2 2 sin(r/2) r
(5.108)
1
Z πy/L sin N+ r
L 2
+ dr.
π 0 r
Define
 1 1
 − , if r 6= 0 and r ∈ [−π, π];
2 sin(r/2) r


g(r) = (5.109)


0, if r = 0.

By virtue of L’Hospital’s Rule, we can see that the function g : [−π, π] → R

defined in (5.109) is continuous and hence bounded on [−π, π]. We can use this
function to rewrite the expression for VN (y) in (5.108) as
L πy/L
Z
1
VN (y) = g(r) sin N + r dr
π 0 2
(5.110)
L (N +1/2)πy/L sin t
Z
+ dt,
π 0 t

1
where we have used the change of variables t = N + r in the second
2
integral on the right–hand side of (5.108).
Next, since
Z (N +1/2)πy/L Z ∞
sin t sin t π
lim dt = dt = ,
N →∞ 0 t 0 t 2
the second integral on the right–hand side of (5.110) is bounded.
We can bound the first integral on the right hand side of (5.110) as follows:
Z πy/L
1
g(r) sin N + r dr 6 π max |g(r)|, for y ∈ [−L, L].
0 2 y∈[−π,π]
Consequently, both integrals on the right–hand side of (5.110) are bounded

independently of N over [−L, L], which establishes the estimate in (5.107).
Next, rewrite (5.105) as
Z δ
[f (x + y) − f (x)]DN (y) dy = [f (x + δ) − f (x)]VN (δ)
−δ
−[f (x − δ) − f (x)]VN (−δ) (5.111)

Z δ
− f 0 (x + y)VN (y) dy,
−δ
for x ∈ [−L, L].

Using the fact that VN is odd (since the Dirichlet kernel is an even function),
we can rewrite the right–hand side of (5.111) as
Z δ
[(f (x + δ) − f (x)) + (f (x − δ) − f (x))]VN (δ) − f 0 (x + y)VN (y) dy. (5.112)
−δ
The first term in (5.112) can be estimated using the Mean Value Theorem and
the estimates in (5.107) and (5.98) as follows:
|[(f (x + δ) − f (x)) + (f (x − δ) − f (x))]VN (δ)| 6 2M M1 δ, (5.113)
for all x ∈ [−L, L].

We can also estimate the second term in (5.112) as follows;
Z δ
− f 0 (x + y)VN (y) dy 6 2M M1 δ, for all x ∈ [−L, L]. (5.114)
−δ
Thus, we can estimate the integral on the left–hand side of (5.111) as

Z δ
[f (x + y) − f (x)]DN (y) dy 6 4M M1 δ, for all x ∈ [−L, L], (5.115)
−δ
where we have used (5.111), (5.112), (5.113), (5.114) and the triangle inequality.
Consequently, the integral I2 in (5.102) can be estimated using (5.115) as
4M M1 δ
|I2 (x)| 6 , for all x ∈ [−L, L]. (5.116)
L
We choose δ > 0 such that
4M M1 δ ε
< ,
L 3
or
εL
δ< .
12M M1
We then obtain from (5.116) that
ε
|I2 (x)| < , for all x ∈ [−L, L]. (5.117)
3
Next, we estimate I3 in (5.103).
Use the formula for the Dirichlet kernel in (5.74) to write (5.103) as
1 L f (x + y) − f (x)
Z
1 πy
I3 (x) = sin N + dy, (5.118)
L δ 2 sin(πy/2L) 2 L
for x ∈ [−L, L].

Define F : [−L, L] × [δ, L] → R by
f (x + y) − f (x)
F (x, y) = , for δ 6 y 6 L and − L 6 x 6 L. (5.119)
2 sin(πy/2L)
Observe that the function F defined in (5.119) is a C 1 function over the rectangle
[−L, L] × [δ, L]. Hence, there exist positive constants M2 and M3 such that
|F (x, y)| 6 M2 , for all δ 6 y 6 L and − L 6 x 6 L, (5.120)

and
∂
[F (x, y)] 6 M3 , for all δ 6 y 6 L and − L 6 x 6 L. (5.121)
∂y
With the definition of F given in (5.119), we can rewrite (5.118) as
1 L
Z
1 πy
I3 (x) = F (x, y) sin N + dy, for x ∈ [−L, L]. (5.122)
L δ 2 L
Next, let
1 π
N+ =λ (5.123)
2 L
in (5.122) to get
Z L
1
I3 (x) = F (x, y) sin λy dy, for x ∈ [−L, L]. (5.124)
L δ
and observe from (5.123 that λ → ∞ as N → ∞.

As in (5.88 in the proof of Proposition 5.1.6, we can use integration by parts
to obtain the estimate
Z L
2M2 M3 L
F (x, y) sin(λx) dx 6 + , for all x ∈ [−L, L]. (5.125)
δ |λ| |λ|
and for λ 6= 0, where we have used the estimates in (5.120) and (5.121).
Combining (5.124), (5.123) and the estimate in (5.125) we obtain
2M2 + M3 L
|I3 (x)| 6 , for all x ∈ [−L, L]. (5.126)
π(N + 1/2)
It follows from (5.126) that, if
6M2 + 3M3 L
N> ,
πε
then
ε
|I3 (x)| <
, for all x ∈ [−L, L].
3
Thus, there exists N1 ∈ N such that
ε
N > N1 ⇒ |I3 (x)| < , for all x ∈ [−L, L]. (5.127)
3
Similar calculations show that there exists N2 ∈ N such that
ε
N > N2 ⇒ |I1 (x)| < , for all x ∈ [−L, L]. (5.128)
3
Letting Nε = max{N1 , N2 } and combining (5.104), (5.117), (5.127) and (5.128),
we obtain that
N > Nε ⇒ |SN (x) − f (x)| < ε, for all x ∈ [−L, L],
which proves that (SN ) converges uniformly to f as N → ∞.
5.1.3 Differentiability of Fourier Series Expansions

The uniform convergence of the Fourier series for f ∈ C 1 (R, R) proved in the
previous section justifies the term–by–term integration that was used in the
derivation of the Fourier coefficients of f in (5.60), (5.61) and (5.62). That fact
alone, however, is not enough to prove that the Fourier series expansion
N nπx
X nπct
uN (x, t) = bn sin cos , for x ∈ [0, L], t > 0, (5.129)
n=1
L L
N = 1, 2, 3, . . ., where bn , for n ∈ N, are given in (5.62), converges uniformly to

a twice–differentiable function
∞ nπx
X nπct
u(x, t) = bn sin cos , for x ∈ [0, L], t > 0. (5.130)
n=1
L L
The function u : [0, L]×[0, ∞) → R defined in (5.130) is a candidate for a solution

of the vibrating string problem in (5.7), provided we can prove that it is well–
defined and its partial derivatives ut , ux , utt and uxx are well defined continuous
functions. This will require term–by–term differentiation of the series in (5.130),
which in turn requires proving that the partial derivatives of the trigonometric
polynomials in (5.129) converge uniformly in [−L, L] as N → ∞. To achieve
this, we will need to make further assumptions on f : [0, L] → R. Indeed, we
will have to assume that f can be extended to a C 2 function over R that is
odd and 2L–periodic. We will also have to obtain more information on the rate
of decay of |an | and |bb |, where an and bn are the Fourier coefficients of f , as
n → ∞.
By virtue of the Riemann–Lebesgue Lemma (Theorem 5.1.4 on page 91 of
these notes), we have that
lim an = 0 and lim bn = 0. (5.131)

n→∞ n→∞
In this section we will see an alternate proof of the limit facts in (5.131 based
in the fact that
∞
X
(|an | + |bn |) < ∞, (5.132)
n=1
for the Fourier coefficients of a 2L–periodic function f ∈ C 1 ([−L, L]). Observe

that the convergence of the series in (5.132) implies the limit facts in (5.131).
We will prove (5.132) as a consequence of the following fact.
Proposition 5.1.8. Assume that f : R → R is a 2L–periodic function that is

square–integrable over [−L, L]; that is,
Z L
|f (x)|2 dx < ∞. (5.133)
−L
Let an and bn , for n ∈ N, denote the Fourier coefficients of f as given in (5.61)

and (5.62), respectively. Then,
∞
X
|an |2 + |bn |2 < ∞.

(5.134)
n=1
To establish Proposition 5.1.8 we will first need to derive some estimates

about sums of squares of Fourier coefficients of a 2L–periodic function f : R → R
that is also square integrable on [−L, L].
For N = 1, 2, 3, . . ., let
N h
X nπx nπx i
SN (x) = ao + an cos + bn sin , (5.135)
n=1
L L
for x ∈ [−L, L], where an , for n = 0, 1, 2, 3, . . ., and bn , for n ∈ N, are given in

(5.60)–(5.62).
For f : R → R, a 2L–periodic function satisfying (5.133), and SN given in
(5.135), compute
Z L Z L Z L Z L
(SN (x) − f (x))2 dx = (SN (x))2 dx − 2 f (x)SN (x)dx + (f (x))2 dx.
−L −L −L −L
(5.136)
We use the integration facts
Z L nπx
cos dx = 0, for all n ∈ N, (5.137)
−L L
Z L nπx
sin dx = 0, for all n ∈ N, (5.138)
−L L

Z L nπx mπx L,
 if n = m;
cos cos dx = (5.139)
−L L L 
0, if n 6= m,


Z L nπx mπx L,
 if n = m;
sin sin dx = (5.140)
−L L L 
0, if n 6= m,

and
Z L nπx mπx
sin cos dx = 0, for all m, n ∈ N, (5.141)
−L L L
to compute the first two integrals on the right–hand side of (5.136).

We begin with
N h
!
X nπx nπx i
2
(SN (x)) = ao + an cos + bn sin
n=1
L L
N h
!
X mπx mπx i
× ao + am cos + bm sin ,
m=1
L L

N
X mπx
(SN (x))2 = a2o + ao am cos
m=1
L
N
X mπx N
X nπx
+ ao bm sin + ao an cos
m=1
L m=1
L
X N
N X nπx mπx
+ an am cos cos
n=1 m=1
L L
X N
N X nπx mπx
+ an bm cos sin (5.142)
n=1 m=1
L L
N
X nπx
+ ao bn sin
m=1
L
X N
N X nπx mπx
+ am bn sin cos
n=1 m=1
L L
X N
N X nπx mπx
+ bn bm sin sin .
n=1 m=1
L L
Next, integrate on both sides of (5.142) from −L to L and use the integration
facts in (5.137)–(5.141) to obtain that
Z L N
X N
X
(SN (x))2 dx = 2La2o + La2n + Lb2n ,
−L n=1 n=1
or
Z L N
X
2
2La2o a2n + b2n ,

(SN (x)) dx = +L for N = 1, 2, 3, . . . (5.143)
−L n=1
Next, we compute
N h
!
Z L Z L X nπx nπx i
f (x)SN (x) dx = f (x) ao + an cos + bn sin dx,
−L −L n=1
L L

Z L Z L
f (x)SN (x) dx = ao f (x) dx
−L −L
" N
#
Z L X nπx
+ f (x) an cos dx
−L n=1
L
" N
#
Z L X nπx
+ f (x) bn sin dx,
−L n=1
L
or
Z L Z L
f (x)SN (x) dx = ao f (x) dx
−L −L
N
X Z L nπx
+ an f (x) cos dx (5.144)
n=1 −L L
N
X Z L nπx
+ bn f (x) sin dx.
n=1 −L L
Next, use the definitions of the Fourier coefficients of f in (5.60)–(5.62) to

rewrite (5.144) as
Z L N
X N
X
f (x)SN (x) dx = 2La2o + La2n + Lb2n ,
−L n=1 n=1
or
Z L N
X
f (x)SN (x) dx = 2La2o + L a2n + b2n .

(5.145)
−L n=1
to rewrite (5.136) as
N
!
Z L Z L X
2 2
2La2o a2n b2n

(SN (x) − f (x)) dx = (f (x)) dx − +L + , (5.146)
−L −L n=1
for N = 1, 2, 3, . . .
Since the left–hand side of the equation in (5.146) is nonnegative, we obtain

from (5.146) that
N Z L
X 1
2a2o + a2n + b2n 6 (f (x))2 dx,

for N = 1, 2, 3, . . . (5.147)
n=1
L −L
Note that, if f is square–integrable over [−L, L], the left–hand side of (5.147) is
monotone, increasing sequence that is bounded above. Hence, it has a limit as
N → ∞. We therefore have that
∞
1 L
X Z
2 2 2
(f (x))2 dx.

2ao + an + bn 6 (5.148)
n=1
L −L
The inequality in (5.148) is an instance of Bessel’s Inequality.

As a consequence of (5.148), we get that
∞
X
a2n + b2n < ∞,

n=1
for the case in which f is square integrable over [−L, L]. We have therefore
proved Proposition 5.1.8.
Next, assume that f : R → R is differentiable, 2L–periodic and with square–
integrable derivative over [−L, L]. Thus, letting a0n and b0n , for n ∈ N, denote
the Fourier coefficients of f 0 , we obtain from Proposition 5.1.8 applied to f 0 that
∞
X
(a0n )2 + (b0n )2 < ∞.

(5.149)
n=1
Next, use integration by parts to derive the identities

nπ
a0n = bn , for n = 1, 2, 3, . . .
L
and
nπ
b0n = − an , for n = 1, 2, 3, . . . ,
L
from which we get
L 1 0
an = − · b , for n = 1, 2, 3, . . . , (5.150)
π n n
and
L 1 0
· a ,
bn = for n = 1, 2, 3, . . . (5.151)
π n n
Taking absolute values on both sides of (5.150) and (5.151), we obtain that
L 1 0
|an | = · |b |, for n = 1, 2, 3, . . . , (5.152)
π n n
and
L 1 0
|bn | = · |a |, for n = 1, 2, 3, . . . (5.153)
π n n
Then, using the inequality
1 2 1 2
|x||y| 6 x + y , for all x, y ∈ R,
2 2
we obtain from (5.152) and (5.153) that

L 1 0 2
|an | 6 + (bn ) , for n = 1, 2, 3, . . . , (5.154)
2π n2
and
L 1 0 2
|bn | 6 + (an ) , for n = 1, 2, 3, . . . (5.155)
2π n2
Adding the estimates in (5.154) and (5.155) we obtain

L 2 0 2 0 2
|an | + |bn | 6 + (an ) + (b n ) , for n = 1, 2, 3, . . .
2π n2
Consequently,
N
" N N
#
X L X 1 X
0 2 0 2

(|an | + |bn |) 6 2 + (an ) + (bn ) , (5.156)
n=1
2π n=1
n2 n=1
for N = 1, 2, 3, . . .
∞
X 1
Now, since the series converges, it follows from Proposition 5.1.8
n=1
n2
that, in the case that f 0 is square–integrable over [−L, L], the sequences on the
right–hand side of (5.156) converge and
N
" ∞ ∞
#
X L X 1 X
0 2 0 2

(|an | + |bn |) 6 2 + (an ) + (bn ) , (5.157)
n=1
2π n=1
n2 n=1
for N = 1, 2, 3, . . .; consequently, if f 0 is square–integrable over [−L, L], the left–

hand side of (5.157) is monotone, increasing sequence that is bounded above.
Hence, it has a limit as N → ∞. We therefore have that
∞
X
(|an | + |bn |) < ∞,
n=1
which is the claim mane in (5.132). We summarize this result in the following
proposition.
Proposition 5.1.9. Assume that f : R → R is a differentiable, 2L–periodic

function whose derivative, f 0 , is square–integrable over [−L, L]; that is,
Z L
|f 0 (x)|2 dx < ∞. (5.158)
−L

∞
X
(|an | + |bn |) < ∞. (5.159)
n=1
Remark 5.1.10. Note that in Proposition 5.1.9 we do not require that f ∈

C 1 (R, R). The square–integrability condition in (5.158) can still hold true even
if f 0 is not continuous; for instance, f 0 could have a finite number of jump
discontinuities and the condition in (5.158) could still be true.
Next, we will show that the summability condition in (5.159) will imply that
the sequence (SN ) converges uniformly to f over [−L, L] as N → ∞. This
result will be attained as a consequence of the Weierstrass Majorant Theorem
or Weierstrass M–Test.
Theorem 5.1.11 (Weierstrass M–Test). Let (gn ) denote a sequence of func-

tions defined on [a, b]. Suppose that there exist positive numbers Mn , for
n = 1, 2, 3, . . ., for which
|gn (x)| 6 Mn , for all x ∈ [a, b] and all n ∈ N,
and
∞
X
Mn < ∞.
n=1
Then, the series

∞
X
gn (x), for x ∈ [a, b],
n=1
converges absolutely and uniformly on [a, b].
(See, for example, [Rud53, Theorem 7.10, pg. 119] for a proof of the Weier-
strass M–Test).
We will also need the following lemma.
Lemma 5.1.12. Let h : R → R be a continuous function that is also 2π–

periodic. Suppose that all the Fourier coefficients of h are 0. Then, h(x) = 0
for all x ∈ R.
Proof of Lemma 5.1.12: We are assuming that h : R → R continuous and 2π–

periodic and that Z π
h(x) dx = 0, (5.160)
−π
Z π
h(x) cos(nx) dx = 0, for all n = 1, 2, 3, . . . , (5.161)
−π
and Z π
h(x) sin(nx) dx = 0, for all n = 1, 2, 3, . . . . (5.162)
−π
We show that
h(x) = 0, for all x ∈ [−π, π]. (5.163)
Assume by way of contradiction that (5.163) is not true. Then, there exists
xo ∈ (−π, π) such that f (xo ) 6= 0. Without loss of generality, we may assume
that f (xo ) > 0. Then, since h is continuous at xo , there exists δ > 0 such that
h(xo )
h(x) > , for all x ∈ (xo − δ, xo + δ), (5.164)
2
where δ can be chosen small enough so that
[xo − δ, xo + δ] ⊂ (−π, π).
Note that the function g : R → R defined by
g(x) = 1 + cos(xo − x) − cos(δ), for x ∈ R, (5.165)
has the following properties:
g(x) > 1, for x ∈ (xo − δ, xo + δ), (5.166)
and
g(x) 6 1, for − π 6 x 6 xo − δ or xo + δ 6 x 6 π. (5.167)
Furthermore, g(x) is a linear combination of the set {1, cos x, sin x}. To see this,
use the trigonometric identity
cos(A − B) = cos A cos A + sin A sin B, for A, B ∈ R,
to write
g(x) = 1 − cos(δ) + cos(δ) cos x + sin(δ) sin x, for x ∈ R;
so that, g is of the form
g(x) = a + b cos x + c sin x, for x ∈ R, (5.168)
and real numbers a, b and c. Furthermore, using the trigonometric identities

1 1
cos2 x = + cos(2x), for all x ∈ R,
2 2
1 1
sin2 x = − cos(2x), for all x ∈ R,
2 2
and
2 sin x cos x = sin(2x), for all x ∈ R,
2
we can see from (5.166) that (g(x)) is a linear combination of elements from
the set
{1, cos x, sin x, cos(2x), sin(2x)}.
Similarly, noting that (g(x))2 = g(x)(g(x))2 and using the trigonometric iden-
tities
1 1
cos x cos(2x) = cos x + cos(3x), for all x ∈ R,
2 2
1 1
sin x sin(2x) = cos x1 cos(3x), for all x ∈ R,
2 2
1 1
cos x sin(2x) = sin x + sin(3x), for all x ∈ R,
2 2
and
1 1
sin x cos(2x) = − sin x + sin(3x), for all x ∈ R,
2 2
we can see that (g(x))3 is a linear combination of of elements from the set
{1, cos x, sin x, cos(2x), sin(2x), cos(3x), sin(3x)}.
Proceeding by induction, and using the appropriate trigonometric identities, we

can show that (g(x))n is span of the set
{1, cos x, sin x, cos(2x), sin(2x), cos(3x), sin(3x), cos(4x), sin(4x), . . .},
for all n ∈ N. (Recall that the span of a set is the collection of all finite a linear
combination of of elements from the set).
It then follows from the assumptions in (5.160), (5.161) and (5.162) that
Z π
h(x)(g(x))n dx = 0, for all n ∈ N. (5.169)
−π
On the other hand, we can estimate the integral in (5.169) by writing

Z π Z xo −δ
h(x)(g(x))n dx = h(x)(g(x))n dx
−π −π
Z xo +δ
+ h(x)(g(x))n dx (5.170)
xo −δ
Z π
+ h(x)(g(x))n dx,
xo +δ
for all n ∈ N, and estimating the first and last integral in the right–hand side
of (5.170) as follows.
Use (5.167) to estimate

Z π Z π
h(x)(g(x))n dx 6 |h(x)| dx, for all n ∈ N;
xo +δ xo +δ
so that Z π
h(x)(g(x))n dx 6 2M π, for all n ∈ N, (5.171)
xo +δ
where
M = max |h(x)| dx. (5.172)
x∈[−π,π]
Similarly,
Z xo −δ
h(x)(g(x))n dx 6 2M π, for all n ∈ N, (5.173)
−π
where M is given in (5.172).

Combining (5.170), (5.171) and (5.173), we then get that
Z π Z xo +δ
h(x)(g(x))n dx > h(x)(g(x))n dx − 4M π, for all n ∈ N, (5.174)
−π xo −δ
where M is given in (5.172).

Next, use the estimates in (5.164) and (5.166) to obtain
Z xo +δ Z xo +δ
h(xo )
h(x)(g(x))n dx > (g(x))n dx;
xo −δ 2 xo −δ
so that, using (5.166) again,

Z xo +δ Z xo +δ/2
h(xo )
n
h(x)(g(x)) dx > (g(x))n dx, for all n ∈ N. (5.175)
xo −δ 2 xo −δ/2
Next, let
r= min g(x). (5.176)
xo −δ/26x6xo +δ/2
Then, by virtue of (5.166),

r > 1. (5.177)
Now, it follows from (5.176) and (5.175) that
Z xo +δ
h(xo ) n
h(x)(g(x))n dx > δr , for all n ∈ N. (5.178)
xo −δ 2
Thus, combining (5.174) and (5.178),

Z π
h(xo )δ n
h(x)(g(x))n dx > r − 4M π, for all n ∈ N. (5.179)
−π 2
Now, since rn → ∞ as n → ∞, by virtue of (5.177), it follows from (5.179) that

it is possible to find an m ∈ N sufficiently large so that
Z π
h(x)(g(x))m dx > 0. (5.180)
−π
We can now see that (5.180) is in direct contradiction with (5.169). Conse-
quently, we conclude that h(x) = 0 for all x ∈ [−π, π], if it is continuous and all
its Fourier coefficients are 0.
Remark 5.1.13. Assume that f : R → R is continuous and 2L–periodic. Sup-

pose that all the Fourier coefficients of f given in (5.60), (5.61) and (5.62) are
zero; so that,
Z L
f (y) dy = 0, (5.181)
−L
Z L nπy
f (y) cos dy = 0, for all n ∈ N, (5.182)
−L L
and Z L nπy
f (y) sin dx = 0, for all n ∈ N. (5.183)
−L L
Define
L
h(x) = f x , for all x ∈ R. (5.184)
π
Then, h : R → R is continuous and 2π–periodic. Furthermore, making the
change of variables
L
y= x (5.185)
π
in the integral in (5.181),
Z π
L L
f x dx = 0,
−π π π
from which we get, using (5.184),

Z π
h(x) dx = 0.
−π
Similarly, making the change of variables give in (5.185) in the integrals in

(5.182) and (5.183), we get that
Z π
h(x) cos(nx) dx = 0, for all n ∈ N,
−π
and Z π
h(x) sin(nx) dx = 0, for all n ∈ N,
−π
respectively.
We have therefore shown that the hypotheses in Lemma 5.1.12 are satisfied
for h given in (5.184), from which we obtain that f (y) = 0 for all y ∈ R.
Theorem 5.1.14 (Uniform Convergence 2). Let f : R → R be a differentiable,

2L–periodic function whose derivative, f 0 , is square–integrable over [−L, L].
Then, the sequence of functions (SN ) given in (5.135) converges absolutely and
uniformly to f over [−L, L].
Proof: Write, using (5.135),

N
X
SN (x) = ao + gn (x), for x ∈ [−L, L], (5.186)
n=1
where
nπx nπx
gn (x) = an cos + bn sin , for x ∈ [−L, L] and n ∈ N. (5.187)
L L
Then, by the triangle inequality, we obtain from (5.187) that
|gn (x)| 6 |an | + |bn |, for x ∈ [−L, L] and n ∈ N.
Thus, by virtue of (5.159) in Proposition 5.1.9, we can apply the Weierstrass

M-Test to conclude that the series
∞
X
gn (x), for x ∈ [−L, L],
n=1
converges absolutely and uniformly to a continuous function, which we shall

denote by g; so that,
∞
X
g(x) = gn (x), for x ∈ [−L, L]. (5.188)
n=1
lim SN (x) = ao + g(x), for all x ∈ [−L, L] uniformly, (5.189)

N →∞
where, according to (5.188) and (5.187),

∞ h
X nπx nπx i
g(x) = an cos + bn sin , for x ∈ [−L, L], (5.190)
n=1
L L
where the series on the right–hand side of (5.190) converges absolutely and
uniformly on [−L, L]. Consequently, the Fourier coefficients of g are an and bn ,
for n ∈ N, which are the same as the Fourier coefficients of f for n ∈ N. Observe
also that Z L
g(x) dx = 0. (5.191)
−L
Consider the function
H(x) = f (x) − ao − g(x), for x ∈ R. (5.192)
It follows from (5.191) and (5.192) that

Z L Z L
H(x) dx = f (x) dx − ao 2L = 0, (5.193)
−L −L
where we have used the definition of ao in (5.60). nπx

Similarly, multiplying on both sides of (5.192) by cos , for n ∈ N,
L
and integrating from −L to L,
Z L nπx Z L nπx Z L nπx
H(x) cos dx = f (x) cos dx − g(x) cos dx,
−L L −L L −L L
nπx
where we have used the fact that cos has mean–value 0, for n ∈ N;
L
consequently, since f and g have the same Fourier coefficients for n ∈ N,
Z L nπx
H(x) cos dx = 0, for n ∈ N. (5.194)
−L L
In the same way we get that

Z L nπx
H(x) sin dx = 0, for n ∈ N. (5.195)
−L L
In view of (5.193), (5.194) and (5.195), we see that the function H : R → R

is a continuous, 2L–periodic function satisfying the conditions of Lemma 5.1.12
(see also Remark 5.1.13 following the proof of Lemma 5.1.12). Consequently,
f (x) − ao − g(x) = 0, for x ∈ R,
f (x) = ao + g(x), for x ∈ R. (5.196)
Combining (5.189) with (5.196) we get that
lim SN (x) = f (x), for all x ∈ [−L, L] uniformly,

N →∞
which was to be shown.

Remark 5.1.15. For the case in which f ∈ C 1 (R, R) is 2L–periodic, we have

that f 0 is bounded on [−L, L]. Consequently, f 0 is square–integrable on [−L, L].
Thus, Theorem 5.1.14 applies in this case, and we therefore recover Theorem
5.1.7.
Example 5.1.16 (Constructing a Solution of the Vibrating String Problem).
Assume that f : [0, L] → R is a differentiable function satisfying f (0) = f (L) =
0. Extend f to an odd, 2L–periodic function, and suppose the extension is
differentiable with square–integrable derivative, f 0 , on the interval [−L, L]. It
then follows from Proposition 5.1.9 that
∞
X
|bn | < ∞, (5.197)
n=1
where Z L
2 nπx
bn = f (x) sin dx, for n ∈ N. (5.198)
L 0 L
We can then use the Weierstrass M-Test (Theorem 5.1.11) to deduce that the
series
∞ nπx
X nπct
bn sin cos , (5.199)
n=1
L L
where bn , for n ∈ N are given in (5.198), converges absolutely and uniformly for
x ∈ [0, L] and t > 0. Indeed, setting
nπx
nπct
gn (x, t) = bn sin cos , for x ∈ [0, L], t > 0 (5.200)
L L
and n ∈ N, we see that
|gn (x, t)| 6 |bn |, for x ∈ [0, L], t > 0 and n ∈ N.
Consequently, in view of (5.197), the Weierstrass M–Test applies. We therefore
deduce that
∞
X
gn (x, t) (5.201)
n=1
converges absolutely and uniformly for x ∈ [0, L] and t > 0. Thus, in view of
(5.200) and (5.201), we see that the series in (5.199) converges absolutely and
uniformly for x ∈ [0, L] and t > 0.
Define
∞ nπx
X nπct
u(x, t) = bn sin cos , for x ∈ [0, L], t > 0. (5.202)
n=1
L L
We have therefore shown that the function u : [0, L]×[0, ∞) → R given in (5.202)
is well–defined. In particular, we get from (5.202) that
∞
X nπx
u(x, 0) = bn sin , for x ∈ [0, L]. (5.203)
n=1
L
It follows from (5.198) and Theorem 5.1.14 that the series on the right–hand
side of (5.203) converges to f (x), for x ∈ [0, L]. Hence,
u(x, 0) = f (x), for x ∈ [0, L]. (5.204)
This is the first of the initial conditions in the vibrating string problem in (5.7).
We note also that the function u defined in (5.202) also satisfies the boundary
conditions in (5.7):
u(0, t) = u(L, t) = 0, for all t > 0.
To see whether or not u in (5.202) satisfies the second initial condition in problem
(5.7), we first need to see that u is differentiable with respect to t and that
∞
∂ X πc nπx nπct
[u(x, t)] = − nbn sin sin , (5.205)
∂t n=1
L L L
for x ∈ [0, L], t > 0. This will require to determine conditions on f for which
that series
∞ nπx
X nπct
nbn sin sin (5.206)
n=1
L L
converges for x ∈ [0, L] and t > 0. We will answer these questions in a subse-
quent example.
We now turn to the question of convergence of the series in (5.206) discussed

in Example 5.1.16. We first note that, if
∞
X
n|bn | < ∞, (5.207)
n=1
the Weierstrass M-Test would imply the absolute and uniform convergence of
the series in (5.206)
Consider the general situation of a 2L–periodic function, f : R → R. Assume
now that f is twice differentiable, with second derivative, f 00 , that is square–
integrable on [−L, L]; so, that
Z L
|f 00 (x)|2 dx < ∞. (5.208)
−L
Let a0n and b0n , for n ∈ N, denote the Fourier coefficients of f 0 . Applying the
result of Proposition 5.1.9 to f 0 , instead of f , we obtain that
∞
X
(|a0n | + |b0n |) < ∞, (5.209)
n=1
in view of (5.208).
We have already seen the identities

nπ
a0n = bn , for n = 1, 2, 3, . . . (5.210)
L
and
nπ
b0n = − an , for n = 1, 2, 3, . . . , (5.211)
L
L
nan = − b0n , for n = 1, 2, 3, . . . , (5.212)
π
and
L 0
nbn = a , for n = 1, 2, 3, . . . . (5.213)
π n
L 0
n(|an | + |bn |) = (|a | + b0n ), for n = 1, 2, 3, . . . . (5.214)
π n
Comparing (5.209) and (5.214) we then see that
∞
X
n(|an | + |bn |) < ∞.
n=1
We have therefore established the following proposition.

Proposition 5.1.17. Assume that f : R → R is a twice–differentiable, 2L–
periodic function whose second derivative, f 00 , is square–integrable over [−L, L];
that is,
Z L
|f 00 (x)|2 dx < ∞.
−L

∞
X
n (|an | + |bn |) < ∞. (5.215)
n=1
Example 5.1.18 (Constructing a Solution of the Vibrating String Problem

(Part II)). In Example 5.1.16 we considered the function u : [0, L] × [0, ∞) → R
defined by
∞ nπx
X nπct
u(x, t) = bn sin cos , for x ∈ [0, L], t > 0, (5.216)
n=1
L L
where the coefficients bn , for n ∈ N are given by
2 L
Z nπx
bn = f (x) sin dx, for n ∈ N, (5.217)
L 0 L
the Fourier coefficients the odd, 2L–periodic extension of f : [0, L] → R.

In Example 5.1.16 we showed that if the odd, 2L–periodic extension, f : R →
R is differentiable with derivative f 0 that is square–integrable over [−L, L], then
the series defining u in (5.216) converges absolutely and uniformly for x ∈
[−L, L] and t > 0.
In this example, we assume further that f : R → R is twice–differentiable
and that the second derivative, f 00 , is square integrable over [−L, L]. We show
∂u ∂u
that the partial derivatives, and , of u exist and are given by
∂t ∂x
∞
∂ X πc nπx nπct
[u(x, t)] = − nbn sin sin , (5.218)
∂t n=1
L L L
and
∞
∂ X π nπx nπct
[u(x, t)] = nbn cos cos , (5.219)
∂x n=1
L L L
respectively.
The series in (5.218) and (5.219) converge absolutely and uniformly, by the
Weierstrass M–Test, provided that we can show that
∞
X
n|bn | < ∞. (5.220)
n=1
However, (5.220) follows from (5.215) in Proposition 5.1.17 because we are as-
suming that f 00 is square integrable over [−L, L]. We therefore conclude that
the series in (5.218) and (5.219) converge absolutely and uniformly for x ∈ [0, L]
and t > 0. In particular, we obtain from (5.218) that
ut (x, 0) = 0, for all x ∈ [0, L;
so that the function u defined in (5.216) satisfies the second of the initial con-
ditions in the vibrating string problem in (5.7).
5.1.4 Solution of the Vibrating String Problem

To complete the construction of a solution of the Vibrating String Problem (5.7)
begun in Example 5.1.16 and Example 5.1.18, we need to see that the function
u : [0, L] × [0, ∞) → R defined in (5.216) has second partial derivatives with
∂2u ∂2u
respect to t and with respect to x, and , for x ∈ (0, L) and t > 0.
∂t2 ∂x2
We have already seen in Example 5.1.18 that if the odd, 2L–periodic exten-
sion, of f : [0, L] → R is twice–differentiable with second–derivative, f 00 , that
∂u ∂u
is square–integrable on [−L, L], then the partial derivatives, and , of
∂t ∂x
u exist and are given by (5.218) and (5.219), respectively. In this section we
will show that, if the odd, 2L–periodic extension of f : [0, L] → R is thrice–
differentiable with third derivative, f 000 , that is square–integrable over [−L, L],
the the function u defined in (5.216) has second partial derivatives, utt and uxx ,
given by
∞
π 2 c2 2 nπx
X nπct
utt (x, t) = − n bn sin cos , (5.221)
n=1
L2 L L
for x ∈ (0, L), t > 0, and

∞
π2 2 nπx
X nπct
uxx (x, t) = − n bn sin cos , (5.222)
n=1
L2 L L
for x ∈ (0, L), t > 0, respectively.

We will be able to apply the Weierstrass M–Test to show that the series in
(5.221) and (5.222) converge absolutely and uniformly, provided that we can
show that
X∞
n2 |bn | < ∞. (5.223)
n=1
The assertion in (5.223) will follow from the following proposition.
Proposition 5.1.19. Assume that f : R → R is a thrice–differentiable, 2L–

periodic function whose third derivative, f 000 , is square–integrable over [−L, L];
that is,
Z L
|f 000 (x)|2 dx < ∞. (5.224)
−L

∞
X
n2 (|an | + |bn |) < ∞. (5.225)
n=1
Proof: Let an , bn , for n ∈ N, denote the Fourier coefficients of f as given in

(5.61) and (5.62. Let a0n , b0n , for n ∈ N, denote the corresponding Fourier
coefficients of f 0 , and a00n , b00n , for n ∈ N, be those corresponding to f 00 .
Applying the result of Proposition 5.1.9 to f 00 , instead of f , we obtain that
∞
X
(|a00n | + |b00n |) < ∞, (5.226)
n=1
in view of (5.224).
Next, use the identities in (5.210) and (5.211), applied to a00n and b00n , to get
that
nπ 0
a00n = b , for n = 1, 2, 3, . . . (5.227)
L n
and
nπ
b00n = − a0n , for n = 1, 2, 3, . . . (5.228)
L
Using the identities in (5.210) and (5.211) again, we obtain from (5.227) and
(5.228) that
π2
a00n = − 2 n2 an , for n = 1, 2, 3, . . .
L
and
π2
b00n = − 2 n2 bn , for n = 1, 2, 3, . . . ,
L
from which we get
π2 2
|a00n | = n |an |, for n = 1, 2, 3, . . . (5.229)
L2
and
π2 2
|b00n | =
n |bn |, for n = 1, 2, 3, . . . . (5.230)
L2
L2
n2 (|an | + |bn |) = (|a00n | + |b00n |) , for n = 1, 2, 3, . . . (5.231)
π2
In view of (5.231) we see that the assertion in (5.225) follows from (5.226), and
the proof of the proposition is now complete.
Theorem 5.1.20 (Existence of Solution for the Vibrating String Problem).

Suppose that f : [0, L] → R satisfies f (0) = f (L) = 0 and that it extends
to an odd, 2L–periodic function, that is thrice–differentiable, and whose third
derivative, f 000 , is square–integrable over [−L, L]. Then, the initial–boundary
value problem


 utt − c2 uxx = 0, for x ∈ (0, L) and t > 0;

u(0, t) = u(L, t) = 0, for t > 0;
(5.232)


 u(x, 0) = f (x), for x ∈ [0, L];
ut (x, 0) = 0, for x ∈ [0, L],

has a solution.
Proof: Define u : [0, L] → R by
∞ nπx
X nπct
u(x, t) = bn sin cos , for x ∈ [0, L], t > 0, (5.233)
n=1
L L
where the coefficients bn , for n ∈ N are given by
2 L
Z nπx
bn = f (x) sin dx, for n ∈ N, (5.234)
L 0 L
the Fourier coefficients the odd, 2L–periodic extension of f : [0, L] → R.
5.2. FUNDAMENTAL SOLUTIONS 119
It follows from Proposition 5.1.19 that

∞
X
n2 |bn | < ∞,
n=1
where the Fourier coefficients of f , bn , for n ∈ N, are given in (5.234). Conse-

quently, by the Weierstrass M–Test, the series on the right–hand side of (5.221)
and (5.222) converge absolutely and uniformly for x ∈ [0, L] and t > 0. Thus,
the function u : [0, L] → R defined in (5.233) has continuous second partial
derivatives, utt and uxx , given in (5.221) and (5.222), respectively. Observe
that
utt (x, t) = c2 uxx (x, t), for x ∈ (0, L) and t > 0.
Thus, the function u given in (5.233) solves the PDE in the Vibrating String
Problem (5.232).
We have already seen in Example 5.1.16 and Example 5.1.18 that the func-
tion u given in (5.233) also satisfies the boundary conditions and the initial
conditions in problem (5.232). Hence, the proof of existence of a solution of
problem (5.232) is now complete.
5.2 Fundamental Solutions

We will illustrate the concept of a fundamental solution by first finding a special
solution of the one–dimensional diffusion equation.
5.2.1 Fundamental Solution to the Diffusion Equation

We compute a very special solution to the one-dimensional diffusion equation
∂u ∂2u
= D 2, for x ∈ R and t > 0. (5.235)
∂t ∂x
In Section 4.2.3 we derived the following dilation–invariant solution to the dif-
fusion equation in (5.235):
√
Z x/ t
2
u(x, t) = c1 e−z /4D
dz + c2 , for x ∈ R and t > 0, (5.236)
0
and constants c1 and c2 . Observe that the function u : R×(0, ∞) → R defined in

(5.236) is a composition of C ∞ functions. It then follows by the Fundamental
Theorem of Calculus and the Chain Rule that u ∈ C ∞ (R × (0, ∞)). Hence,
we can differentiate on both sides of the PDE in (5.235) with respect to x, for
example, and get the valid statement
utx = D uxxx ;
thus, by the equality of the mixed partial derivatives,
(ux )t = D (ux )xx ,
which shows that ux is also a solution of the one–dimensional diffusion equation

in (5.235). Hence, by taking the partial derivative with respect to x in (5.236)
we obtain another solution to the the one–dimensional diffusion equation in
(5.235). Set v(x, t) = ux (x, t), for x ∈ R and t > 0, where u is given in (5.236).
Then, using the Fundamental Theorem of Calculus and the Chain Rule, we
obtain from (5.236) that
c1 2
v(x, t) = √ e−x /4Dt , for x ∈ R and t > 0, (5.237)
t
and some constant c1 , is a solution to the one–dimensional diffusion equation

in (5.235).
Z An interesting property of the function defined in (5.237) is that the integral
∞
v(x, t) dx is finite and is independent of t > 0. Indeed, using the fact that
−∞
Z ∞ √
2
e−z dz = π,
−∞
and making the change of variables

x
z=√ ,
4Dt
so that √
dx = 4Dt dz,
we obtain, for t > 0,
∞
c1 √ ∞
Z Z
2
v(x, t) dx = √ 4Dt e−z dz,
−∞ t −∞
or Z ∞ √
v(x, t) dx = c1 4Dπ, for all t > 0. (5.238)
−∞
We chose the constant c1 in (5.238) so that

Z ∞
v(x, t) dx = 1, for all t > 0;
−∞
that is,
1
c1 = √ . (5.239)
4Dπ
Substituting the value of c1 in (5.239) into the definition of v(x, t) in (5.237),
we obtain
1 2
v(x, t) = √ e−x /4Dt , for x ∈ R and t > 0. (5.240)
4πDt
We shall denote the expression for v(x, t) defined in (5.240) by p(x, t), so that
1 2
p(x, t) = √ e−x /4Dt , for x ∈ R and t > 0. (5.241)
4πDt
It then follows from what we have shown thus far that the function p defined in
(5.241) is a C ∞ function defined in R × (0, ∞) that solves the one–dimensional
diffusion equation in (5.235); that is,
∂p ∂2p
= D 2, for x ∈ R and t > 0. (5.242)
∂t ∂x
Also, it follows from (5.238) and (5.239) that
Z ∞
p(x, t) dx = 1, for all t > 0. (5.243)
−∞
In fact, using a change of variables we obtain from (5.243) that

Z ∞
p(x − y, t) dy = 1, for all x ∈ R and t > 0. (5.244)
−∞
In addition to (5.244), the function p defined in (5.241) has the following prop-
erties:
Proposition 5.2.1 (Properties of p). Let p(x, t) be as defined in (5.241) for
x ∈ R and t > 0.
(i) p(x − y, t) > 0 for all x, y ∈ R and t > 0
(ii) If x 6= y, then lim p(x − y, t) = 0.
t→0+
(iii) If x = y, then lim+ p(x − y, t) = +∞.

t→0
See Problem 5 in Assignment #14.

In this section we will see how to use the properties in (5.244) and in
Proposition 5.2.1 to obtain a solution to the initial value problem for the one–
dimensional diffusion equation
∂u ∂2u


 = D 2, x ∈ R, t > 0;
∂t ∂x (5.245)


u(x, 0) = f (x), x ∈ R,
where f : R → R is a bounded function that is also piecewise continuous.

Definition 5.2.2 (Piecewise Continuous Functions). A function f : R → R is
said to have a jump discontinuity at x ∈ R if the one–sided limits
lim f (y) and lim f (y)

y→x+ y→x−
exist and
lim f (y) 6= lim− f (y).
y→x+ y→x
We say that f is piecewise continuous if it is continuous except at an at most

countable number of points at which f has jump discontinuities.
Figure 5.2.1 shows a portion of the sketch of a piecewise continuous function.

We will show that the function u : R × (0, ∞) → R given by
Figure 5.2.1: Sketch of a Piecewise Continuous Function
Z ∞
u(x, t) = p(x − y, t)f (y) dy, for x ∈ R and t > 0, (5.246)
−∞
is a candidate for a solution of the initial value problem in (5.245). We note

that, since p(x − y, t)) is not defined at t = 0, the initial condition in the IVP
in (5.245) has to be understood as
lim u(x, t) = f (x).

t→0+
We will see in this section that (5.247) holds true foe values of x at which f is
continuous. For values of x at which f has a jump discontinuity
f (x+ ) + f (x− )
lim u(x, t) = ,
t→0+ 2
where f (x+ ) and f (x− ) are the one–sided limits
f (x+ ) = lim+ f (y) and f (x− ) = lim− f (y),

y→x y→x
respectively.
We state the main result of this section as the following proposition:
Proposition 5.2.3. Let u be given by (5.246), where f : R → R is a bounded,

piecewise continuous function. Then, u is C 2,1 (R × (0, ∞))1 and
∂u ∂2u
(x, t) = D 2 (x, t), for x ∈ R and t > 0. (5.247)
∂t ∂x
Furthermore,
lim u(x, t) = f (x). (5.248)
t→0+
if f is continuous at x, and
f (x+ ) + f (x− )
lim u(x, t) = , (5.249)
t→0+ 2
if f has a jump discontinuity at x.
Once we have proved Proposition 5.2.3, we will have constructed a solution
Z ∞
−∞
to the initial value problem of the initial value for the one–dimensional diffusion
for the case of continuous initial data f , where p is defined in (5.241). Thus,
a solution of the initial value problem in (5.245) is obtained by integrating
f (y)p(x − y, t) over y in the entire real line. The map
(x, y, t) 7→ p(x − y, t), for all x, y ∈ R and t > 0,
or
1 2
(x, y, t) 7→ √e−(x−y) /4Dt , for all x, y ∈ R and t > 0,
4πDt
is usually called the heat kernel; we shall also call it the fundamental solu-
tion to the one–dimensional diffusion equation. We will denote it by K(x, y, t),
so that K : R2 × (0, ∞) → R and
1 2
K(x, y, t) = √ e−(x−y) /4Dt , for all x, y ∈ R and t > 0. (5.251)
4πDt
We shall reiterate the properties of the heat kernel that we have discussed for
future reference in the following proposition, we will add the additional obser-
vation that K is symmetric in x and y; that is K(x, y, t) = K(y, x, t) for all
x, y ∈ R and t > 0.
Proposition 5.2.4 (Properties of the Heat Kernel). Let K(x, y, t) be as defined
in (5.251) for x, y ∈ R and t > 0.
(i) K(x, y, t) = K(y, x, t) for all x, y ∈ R and t > 0.
(ii) K(x, y, t) > 0 for all x, y ∈ R and t > 0.
1 The function u is C 2 in the first variable, and C 1 in the second variable
Z ∞
(iii) K(y, x, t) dy = 1 for all x ∈ R and t > 0.
−∞
(iv) If x 6= y, then lim+ K(x, y, t) = 0.

t→0
(v) If x = y, then lim+ K(x, y, t) = +∞.

t→0
Before we prove Proposition 5.2.3, we will establish two Lemmas; the first
one involves the error function,
Erf : R → R,
defined by Z x
2 2
Erf(x) = √ e−r dr, for x ∈ R, (5.252)
π 0
and its properties:
Proposition 5.2.5. Let Erf : R → R be as given in (5.252). Then,
(i) Erf(0) = 0;
(ii) lim Erf(x) = 1;
x→∞
(iii) lim Erf(x) = −1;

x→−∞
See Problem 1 in Assignment #14.

A sketch of the graph of y = Erf(x) is shown in Figure 5.2.2.
Figure 5.2.2: Sketch of Graph of Error Function
Lemma 5.2.6. Let p(x, t) be as defined in (5.241) for x ∈ R and t > 0. For
δ > 0, Z ∞
lim p(x, t) dx = 0. (5.253)
t→0+ δ
and Z −δ
lim+ p(x, t) dx = 0. (5.254)
t→0 −∞
x
Proof: Make the change of variables y = √ to write
4Dt
Z ∞ Z ∞
1 2
p(x, t) dx = √ · e−x /4Dt dx
δ δ 4πDt
Z ∞
1 2
= √ √ e−y dy
π δ/ 4Dt

1 δ
= 1 − Erf √ ,
2 4Dt
where we have used the definition of the error function in (5.252) and the fact
that Z ∞ √
2 π
e−y dy = .
0 2
We then have that
Z ∞
1 δ
p(x, t) dx = 1 − Erf √ , for t > 0. (5.255)
δ 2 4Dt
Now, it follows from (5.255) and (ii) in Proposition 5.2.5 that

Z ∞
lim p(x, t) dx = 0,
t→0+ δ
which is (5.253). Similar calculations can be used to derive (5.254).
Lemma 5.2.7. Let p(x, t) be as defined in (5.241) for x ∈ R and t > 0. Then, we
have the following estimates on integrals of the absolute values of the derivatives
of p:
Z ∞
∂p 1
(x − y, t) dy 6 , for all x ∈ R and t > 0, (5.256)
−∞ ∂t t
and
Z ∞
∂p 1
(x − y, t) dy = √ , for all x ∈ R and t > 0, (5.257)
−∞ ∂x πDt
Proof: Compute the partial derivative of
1 2
p(x − y, t) = √ e−(x−y) /4Dt , for all x, y ∈ R and t > 0, (5.258)
4πDt
with respect to t to obtain
∂ 1 (x − y)2
[p(x − y, t)] = − p(x − y, t) + p(x − y, t), (5.259)
∂t 2t 4Dt2
for all x, y ∈ R and t > 0. Next, take absolute value on both sides of (5.259),
apply the triangle inequality, and use the positivity of the heat kernel (see (ii)
in Proposition 5.2.4) to get
∂ 1 (x − y)2
[p(x − y, t)] 6 p(x − y, t) + p(x − y, t), (5.260)
∂t 2t 4Dt2
for all x, y ∈ R and t > 0. Integrating on both sides of (5.260) and using (5.244)
(see (iii) in Proposition 5.2.4) yields
Z ∞ Z ∞
∂ 1 (x − y)2
[p(x − y, t)] dy 6 + 2
p(x − y, t) dy, (5.261)
−∞ ∂t 2t −∞ 4Dt
for all x ∈ R and t > 0.

Next, we evaluate the right–most integral in (5.261),
∞ ∞ 2
(x − y)2 (x − y)2 e−(x−y) /4Dt
Z Z
p(x − y, t) dy = · √ dy,
−∞ 4Dt2 −∞ 4Dt2 4πDt
by making the change of variables
y−x
ξ=√ ,
4Dt
so that ∞ ∞
(x − y)2
Z Z
1 2
p(x − y, t) dy = √ ξ 2 e−ξ dξ, (5.262)
−∞ 4Dt2 t π −∞
for all x ∈ R and t > 0. The right–most integral in (5.262) can be evaluated
using integration by parts to yield
Z ∞ Z ∞
2 −ξ 2 2
ξ e dξ = 2 ξ 2 e−ξ dξ
−∞ 0
∞
Z ∞
−ξ 2 2
= −ξe + e−ξ dξ,
0 0
so that Z ∞ √
2 π
ξ 2 e−ξ dξ = . (5.263)
−∞ 2
Combining (5.263), (5.262) and (5.261) yields the estimate
Z ∞
∂ 1
[p(x − y, t)] dy 6 , for x ∈ R and t > 0,
−∞ ∂t t
which is (5.256).
In order to establish (5.257), first take the partial derivative with respect to
x on both side of (5.258) to get
∂ x−y
[p(x − y, t)] = − p(x − y, t), for x ∈ R and t > 0, (5.264)
∂x 2Dt
so that, taking absolute value on both sides of (5.264) and integrating,
Z ∞ Z ∞
∂ |x − y|
[p(x − y, t)] dy = p(x − y, t) dy, (5.265)
−∞ ∂x −∞ 2Dt
Evaluate the right–most integral in (5.265),

∞ ∞ 2
|x − y| |x − y| e−(x−y) /4Dt
Z Z
p(x − y, t) dy = · √ dy, (5.266)
−∞ 2Dt −∞ 2Dt 4πDt
by making the change of variables
y−x
ξ=√ ,
4Dt
to get
∞ 2 ∞
|x − y| e−(x−y) /4Dt
Z Z
1 2
· √ dy = √ |ξ|e−ξ dξ
−∞ 2Dt 4πDt πDt −∞
Z ∞
2 2
= √ ξe−ξ dξ,
πDt 0
so that 2
∞
|x − y| e−(x−y) /4Dt
Z
1
· √ dy = √ . (5.267)
−∞ 2Dt 4πDt πDt
The statement in (5.257) now follows by putting together the results in (5.267),
(5.266) and (5.265).
Proof of Proposition 5.2.3: Let f : R → R be a piecewise continuous function

satisfying
|f (x)| 6 M, for all x ∈ R, (5.268)
and some positive constant M , and define u : R × (0, t) → R by
Z ∞
−∞
where p(x − y, t) denotes the heat kernel given in (5.258). We will show that
u solves the one–dimensional diffusion equation in (5.247). Before we do that,
though, we need to verify that the expression in (5.269) does indeed define a
function u : R × (0, ∞) → R. In order to do this we need to make sure that

the integral on the right–hand side of (5.269) is a real number. This will follow
from the estimate
Z ∞
|p(x − y, t)f (y)| dy < ∞ for x ∈ R and t > 0. (5.270)
−∞
In order to derive the estimate in (5.270), use the positivity of the heat kernel
(see (ii) in Proposition 5.2.4), (5.244) and (5.268) to compute
Z ∞ Z ∞
|p(x − y, t)f (y)| dy 6 M p(x − y, t) dy,
−∞ −∞
so that
Z ∞
|p(x − y, t)f (y)| dy 6 M, for x ∈ R and t > 0, (5.271)
−∞
which implies (5.270). Observe that the estimate in (5.271) also implies that
|u(x, t)| 6 M, for x ∈ R and t > 0,
by virtue of (5.269).
The fact that u defined in (5.269) solves the one–dimensional diffusion equa-
tion in (5.247) will follow from the fact that the heat kernel itself solves the
one–dimensional heat equation,
∂ ∂2
[p(x − y, t)] = D 2 [p(x − y, t)], for x, y ∈ R and t > 0; (5.272)
∂t ∂x
(see also (5.242). Indeed, suppose for the moment that we can interchange
differentiation and integration in the definition of u in (5.269), so that
Z ∞
∂u ∂p
(x, t) = (x − y, t)f (y) dy, for x ∈ R and t > 0, (5.273)
∂t −∞ ∂t
and
∞
∂2u ∂2p
Z
(x, t) = (x − y, t)f (y) dy, for x ∈ R and t > 0. (5.274)
∂x2 −∞ ∂x2
Thus, combining (5.273) and (5.274),
Z ∞
∂2u ∂2p

∂u ∂p
(x, t) − D 2 (x, t) = (x − y, t) − D 2 (x − y, t) f (y) dy,
∂t ∂x −∞ ∂t ∂x
which shows that (5.247) holds true by virtue of (5.272)
The expressions in (5.273) and (5.274) are justified by the assumption that
f is bounded (see (5.268) and the estimates (5.256) and (5.257) in Lemma 5.2.7;
namely, Z ∞
∂p 1
(x − y, t) dy 6 , for all x ∈ R and t > 0.
−∞ ∂t t
and Z ∞
∂p 1
(x − y, t) dy = √ , for all x ∈ R and t > 0.
−∞ ∂x πDt
Observe that, (5.274) and (5.272) imply the estimate
Z ∞ 2
∂ p 1
2
(x − y, t) dy 6 , for all x ∈ R and t > 0.
−∞ ∂x Dt
We have therefore established that the function u : R × (0, 1) defined in (5.269)

is a C 2 function in the first variable, C 1 in the second variable, and is a solution
to the one–dimensional diffusion equation.
Next, we will prove the second assertion in Proposition 5.2.3.

(i) Assume first that f is continuous at x and let ε > 0 be given. Then, there
exists δ > 0 such that
ε
|y − x| < δ ⇒ |f (y) − f (x)| < . (5.275)
3
We consider
Z ∞ Z ∞
u(x, t) − f (x) = p(xo − y, t)f (y) dy − f (x) p(xo − y, t) dy,
−∞ −∞
where we have used the definition of u(x, t) in (5.269) and (5.244) (see also the
fact (iii) in Proposition 5.2.4). We then have that
Z ∞
u(x, t) − f (x) = p(x − y, t)(f (y) − f (x)) dy,
−∞
so that Z ∞
|u(x, t) − f (x)| 6 p(x − y, t)|f (y) − f (x)| dy, (5.276)
−∞
where we have used the fact that p(x, t) is positive for all x ∈ R and all t > 0.
Next, re-write the integral on the right–hand side of (5.276) as a sum of
three integrals,
Z ∞
p(x − y, t)|f (y) − f (x)| dy =
−∞
Z x−δ
p(x − y, t)|f (y) − f (x)| dy
−∞
(5.277)
Z x+δ
+ p(x − y, t)|f (y) − f (x)| dy
x−δ
Z ∞
+ p(x − y, t)|f (y) − f (x)| dy.
x+δ
We first estimate the middle integral on the right–hand side of (5.277), using
(5.275) and (5.244) to get
Z x+δ
ε
p(x − y, t)|f (y) − f (x)| dy < . (5.278)
x−δ 3
Next, use (5.268) and the triangle inequality to obtain the following estimate
for the last integral on the right–hand side of (5.277),
Z ∞ Z ∞
p(x − y, t)|f (y) − f (x)| dy 6 2M p(x − y, t) dy. (5.279)
x+δ x+δ
Make the change of variables ξ = y − x in the integral on the right–hand side

of (5.279) to obtain
Z ∞ Z ∞
p(x − y, t)|f (y) − f (xo )| dy 6 2M p(ξ, t) dξ, (5.280)
x+δ δ
where we have also used the symmetry of the heat kernel (see (i) in Proposition
5.2.4). It follows from (5.280) and (5.253) in Lemma 5.2.6 that
Z ∞
lim+ p(x − y, t)|f (y) − f (x)| dy = 0;
t→0 x+δ
thus, there exists δ1 > 0 such that

Z ∞
ε
0 < t < δ1 ⇒ p(x − y, t)|f (y) − f (x)| dy < . (5.281)
x+δ 3
Similar calculations to those leading to (5.281), using (5.254) in Lemma 5.2.6,

can be used to show that there exists δ2 > 0 such that
Z x−δ
ε
0 < t < δ2 ⇒ p(x − y, t)|f (y) − f (x)| dy < . (5.282)
−∞ 3
Let δ3 = min{δ1 , δ2 }. It then follows from (5.277), in conjunction with (5.278),

(5.281) and (5.282), that
Z ∞
0 < t < δ3 ⇒ p(x − y, t)|f (y) − f (xo )| dy < ε.
−∞
We have therefore proved that

Z ∞
lim+ p(x − y, t)|f (y) − f (x)| dy = 0. (5.283)
t→0 −∞
It follows from (5.283) and the estimate in (5.276) that
lim |u(x, t) − f (x)| = 0,

t→0+
which yields (5.248) and assertion (i) of Proposition 5.2.3 has been proved.
(ii) Assume that f has a jump discontinuity at x and put
f (x+ ) = lim+ f (y) and f (x− ) = lim− f (y). (5.284)

y→x y→x
Let ε > 0 be given. It follows from (5.284) that there exists δ > 0 such that
ε
x < y < x + δ ⇒ |f (y) − f (x+ )| < , (5.285)
3
and
ε
x − δ < y < x ⇒ |f (y) − f (x− )| < . (5.286)
3
Use the definition of u(x, t) in (5.269) to write
∞
f (x+ ) + f (x− )
Z
1 1
u(x, t) − = p(x − y, t)f (y) dy − f (x+ ) − f (x− ),
2 −∞ 2 2
and note that

Z xo Z ∞
1
= p(xo − y, t) dy = p(xo − y, t) dy, (5.287)
2 −∞ xo
by virtue of (5.244), (5.243) and the symmetry of the heat kernel (see (i) in
Proposition 5.2.4). We therefore have that
f (x+ ) + f (x− )
u(x, t) −
2
Z x
= p(x − y, t)(f (y) − f (x− )) dy
−∞
Z ∞
+ p(x − y, t)(f (y) − f (x+ )) dy,
xo
so that
f (x+ ) + f (x− )
u(x, t) −
2
Z x
6 p(x − y, t)|f (y) − f (x− )| dy (5.288)
−∞
Z ∞
+ p(x − y, t)|f (y) − f (x+ )| dy,
x
We re-write the last integral on the right–hand side of (5.288) as a sum of two
integrals,
Z ∞
p(x − y, t)|f (y) − f (x+ )| dy
x
Z x+δ
= p(x − y, t)|f (y) − f (x+ )| dy (5.289)
x
Z ∞
+ p(x − y, t)|f (y) − f (x+ )| dy,
x+δ
where
Z x+δ Z x+δ
+ ε ε
p(x − y, t)|f (y) − f (x )| dy < p(x − y, t) dy < , (5.290)
x 3 x 6
by virtue of (5.286) and (5.287).

Similar calculations to those leading to (5.281) can be used to show that
there exists δ1 > 0 such that
Z ∞
ε
0 < t < δ1 ⇒ p(x − y, t)|f (y) − f (x+ )| dy < . (5.291)
x+δ 3
Combining (5.290) and (5.291), we obtain from (5.289) that

Z ∞
ε
0 < t < δ1 ⇒ p(x − y, t)|f (y) − f (x+ )| dy < . (5.292)
x 2
Similarly, we can show that there exists δ2 > 0 such that

Z x
ε
0 < t < δ2 ⇒ p(x − y, t)|f (y) − f (x− )| dy < . (5.293)
−∞ 2
Thus, letting δ3 = min{δ1 , δ2 } we see that the conjunction of (5.292) and (5.293),
together with (5.288), implies that
f (x+ ) + f (x− )
0 < t < δ3 ⇒ u(x, t) − < ε.
2
We have therefore established (5.249) and the proof of part (ii) of Proposition
5.2.3 is now complete.
Example 5.2.8. Solve the initial value problem for the diffusion equation in
(5.245), where (
1, if − 1 < x 6 1;
f (x) = (5.294)
0, elsewhere.
Figure 5.2.3: Initial Condition for Example 5.2.8
Solution: A sketch of the graph of the initial condition, f , is shown in Figure

5.2.3. Note that f has jump discontinuities at −1 and at 1.
Using the formula in (5.269) we get that a solution to the initial value prob-
lem (5.245) with initial condition given in (5.294) is given by
Z 1
u(x, t) = p(x − y, t) dy, for x ∈ R and t > 0,
−1
or
Z 1
1 2
u(x, t) = √ e−(x−y) /4Dt
dy, for x ∈ R and t > 0, (5.295)
4πDt −1
x−y
Make the change variables r = √ in (5.295) to obtain
4Dt
Z x−1
√
1 4Dt 2
u(x, t) = − √ e−r dr, for x ∈ R and t > 0,
π x+1
√
4Dt
or x+1 x−1
Z √ Z √
1 4Dt 2 1 4Dt 2
u(x, t) = √ e−r dr − √ e−r dr, (5.296)
π 0 π 0
for x ∈ R and t > 0.
Making use of the error function defined in (5.252), we can rewrite (5.296)
as
1 x+1 x−1
u(x, t) = Erf √ − Erf √ , (5.297)
2 4Dt 4Dt
for x ∈ R and t > 0. Figure 5.2.4 shows plots of the graph of y = u(x, t),
where u(x, t) is as given in (5.297), for various values of t in the case 4D = 1.
A few interesting properties of the function u given in (5.297) are apparent by
examining the pictures in Figure 5.2.4. First, the graph of y = u(x, t) is smooth
for all t > 0. Even though the initial temperature distribution, f , in (5.294)
is not even continuous, the solution to the initial value problem (5.245) given
in (5.297) is in fact infinitely differentiable as soon as the process gets going
for t > 0. Secondly, the values, u(x, t), of the function u given in (5.297)) are
positive at all values of x ∈ R and t > 0. In particular, for values of x with
|x| > 1, where the initial temperature is zero, the temperature rises instantly
1.0
0.8
0.6
0.5 HerfH3.16228 Hx + 1LL - erfH3.16228

0.4
0.5 HerfH1 - xL + erfHx + 1LL

0.2
x+1 x-1
0.5 erf - erf
-3 -2 -1 1 2 3 10 10
Computed by WolframÈAlpha
Figure 5.2.4: Sketch of Graph of y = u(x, t) for t = 0.1, 1, 10
for t > 0. Thus, the diffusion model for heat propagation predicts that heat
propagates with infinite speed. Thirdly, we see from the pictures in Figure 5.2.4
that
lim u(x, t) = 0, for all x ∈ R. (5.298)
t→∞

5.2.2 Uniqueness for the Diffusion Equation

The observation (5.298) in Example 5.2.8 is true in general for solutions to the
initial value problem in (5.245) for the case in which the initial condition, f , is
square–integrable; that is,
Z ∞
|f (x)|2 dx < ∞ (5.299)
−∞
Observe that, for the function f in Example 5.2.8 satisfies

Z ∞
|f (x)|2 dx = 2,
−∞
so that the integrability condition in (5.299) holds true for the function in
(5.294).
Before we establish that (5.298) is true for any solution of the initial value
problem (5.245) in which the initial condition satisfies (5.299), we will first need
to derive other properties of the function u given in (5.246).
Proposition 5.2.9. Let f : R → R be continuous and satisfying (5.299); that
is, Z ∞
|f (x)|2 dx < ∞.
−∞
Put
Z ∞
u(x, t) = p(x − y, t)f (y) dy, for x ∈ R and t > 0. (5.300)
−∞
Then, Z ∞
|u(x, t)|2 dx < ∞, for all t > 0, (5.301)
−∞
and Z ∞ 2
∂u
(x, t) dx < ∞, for all t > 0. (5.302)
−∞ ∂x
Proof: Let u be given by (5.300), where f satisfies the condition in (5.299).

Apply the Cauchy–Schwarz inequality (or Jensen’s Inequality) to get
Z ∞
2
|u(x, t)| 6 p(x − y, t)|f (y)|2 dy, for x ∈ R and t > 0, (5.303)
−∞
where we have also used (5.244) and the positivity of the heat kernel (see (ii)
and (iii) in Proposition 5.2.4).
Integrate with respect to x on both sides of (5.303) to get
Z ∞ Z ∞Z ∞
|u(x, t)|2 dx 6 p(x − y, t)|f (y)|2 dydx, (5.304)
−∞ −∞ −∞
for t > 0. Interchanging the order of integration in the integral on the right–
hand side of (5.304) we obtain
Z ∞ Z ∞ Z ∞
|u(x, t)|2 dx 6 |f (y)|2 p(x − y, t) dx dy, (5.305)
−∞ −∞ −∞
for t > 0. It follows from (5.305) and (5.244) that

Z ∞ Z ∞
2
|u(x, t)| dx 6 |f (y)|2 dy, for t > 0, (5.306)
−∞ −∞
Combining (5.306) and (5.299) then yields

Z ∞
|u(x, t)|2 dx < ∞, for all t > 0, (5.307)
−∞
which is the condition in (5.301).

Next, differentiate u in (5.300) with respect to x to get
∞ 2
(x − y) e−(x−y) /4Dt
Z
∂u
(x, t) = − √ f (y) dy,
∂x −∞ 2Dt 4πDt
so that ∞
(x − y)
Z
∂u
(x, t) = − p(x − y, t) f (y) dy, (5.308)
∂x −∞ 2Dt
Proceeding as in the first part of this proof, use the Cauchy–Schwarz in-
equality (or Jensen’s inequality) to obtain from (5.308) that
2 ∞
(x − y)2
Z
∂u
(x, t) 6 p(x − y, t) |f (y)|2 dy, (5.309)
∂x −∞ 4D2 t2

Next, integrate on both sides of (5.309) with respect to x and interchange
the order of integration to obtain
Z ∞ 2 Z ∞ Z ∞
∂u 1 2
(x, t) dx 6 |f (y)| (x−y)2 p(x−y, t) dx dy, (5.310)
−∞ ∂x 4D2 t2 −∞ −∞
for t > 0.
Observe that the inner integral in the right–hand side of (5.310) is simply
the variance, 2Dt, of the probability density function p(x, t), so that
Z ∞
(x − y)2 p(x − y, t) dx = 2Dt, for all y ∈ R and t > 0. (5.311)
−∞
Putting together (5.310) and (5.311)

Z ∞ 2 Z ∞
∂u 1
(x, t) dx 6 |f (y)|2 dy, for t > 0,
−∞ ∂x 2Dt −∞
which implies (5.302) by virtue of (5.299).

We will next show that, if in addition to the integrability condition in (5.299)
for the initial distribution, f , we also impose the conditions (5.301) and (5.302)
on the initial value problem (5.245), then any solution must be of the form given
in (5.246). This amounts to showing that the initial value problem (5.245) in
which the initial condition satisfies (5.299), together with the integrability con-
dition in (5.301) and (5.302), has a unique solution. We will need the estimate
in the following lemma when we prove uniqueness.
Lemma 5.2.10. Let f : R → R be a continuous function satisfying (5.299). Let
v be any solution of the problem
∂2u

∂u
= D 2, for x ∈ R, t > 0;





 ∂t ∂x



u(x, 0) = f (x), for x ∈ R;






Z ∞ (5.312)


 |u(x, t)|2 dx < ∞, for all t > 0;
 −∞





 Z ∞ 2

 ∂u
(x, t) dx < ∞, for all t > 0.


−∞ ∂x

Then, Z ∞ Z ∞
|v(x, t)|2 dx 6 |f (x)|2 dx, for t > 0. (5.313)
−∞ −∞
Proof: Let v denote any solution to the problem (5.312), where f satisfies the
integrability condition in (5.299).
In order to establish (5.313), set
Z ∞
E(t) = |v(x, t)|2 dx, for all t > 0. (5.314)
−∞
It follows from the integrability condition in (5.312) that E(t) in (5.314) is well
defined for all t > 0 as a real valued function, E : [0, ∞) → R. Note also that
Z ∞
E(0) = |f (x)|2 dx, (5.315)
−∞
by virtue of the initial condition in problem (5.312).

Next, observe that, since v satisfies the diffusion equation in (5.312), that is
vt = Dvxx ,
then E is differentiable and
Z ∞ Z ∞
0
E (t) = 2v(x, t)vt (x, t) dx = 2D v(x, t)vxx (x, t) dx, (5.316)
−∞ −∞
for t > 0.
We note that the integrability conditions in (5.312) imply that
lim v(x, t) = 0 and lim v(x, t) = 0, for t > 0, (5.317)
x→∞ x→−∞
and
lim vx (x, t) = 0 and lim vx (x, t) = 0. for t > 0, (5.318)
x→∞ x→−∞
Integrate by parts the last integral in (5.316) to get

" #
Z R
0 2
E (t) = lim v(R, t)vx (R, t) − v(−R, t)vx (−R, t) − (vx (x, t)) dx ,
R→∞ −R
so that Z ∞ 2
∂v
E 0 (t) = − (x, t) dx, for t > 0, (5.319)
−∞ ∂x
by virtue of (5.317), (5.318) and the last integrability condition in (5.312).
Now, it follows from (5.319) that
E 0 (t) 6 0, for all t > 0,
so that E is nondecreasing in t and therefore
E(t) 6 E(0), for all t > 0. (5.320)
The estimate in (5.313) follows from (5.320) in view of (5.314) and (5.315).
Proposition 5.2.11. Let f : R → R be a continuous function satisfying (5.299).

The problem
∂2u

∂u
= D 2, for x ∈ R, t > 0;



 ∂t ∂x





u(x, 0) = f (x), for x ∈ R;






Z ∞ (5.321)


 |u(x, t)|2 dx < ∞, for all t > 0;
 −∞





 Z ∞ 2

 ∂u
(x, t) dx < ∞, for all t > 0,


−∞ ∂x

has at most one solution.

Proof: Let v be any solution of the problem in (5.321) and let u be given by
(5.300). It follows from Proposition 5.2.3 and Proposition 5.2.9 that u solves
problem (5.321). Put
w(x, t) = v(x, t) − u(x, t), for x ∈ R and t > 0. (5.322)
It follows from the linearity of the differential equation in (5.321) that w also
solves the diffusion equation; indeed,
wt = vt − ut = Dvxx − Duxx = D(vxx − uxx ) = Dwxx .
The function w defined in (5.322) also satisfies the integrability condition in

problem (5.321) ; in fact, by the triangle inequality,
|w(x, t)| 6 |v(x, t)| + |u(x, t)|,
so that
|w(x, t)|2 6 |v(x, t)|2 + 2|v(x, t)| · |u(x, t)| + |u(x, t)|2 , (5.323)
for all x ∈ R and all t > 0. Next, use the inequality
2ab 6 a2 + b2 , for a, b ∈ R,
in (5.323) to get
|w(x, t)|2 6 2 |v(x, t)|2 + |u(x, t)|2 ,

for x ∈ R and t > 0. (5.324)
Integrating on both sides of (5.324) with respect to x we then obtain that

Z ∞ Z ∞ Z ∞
2 2 2
|w(x, t)| dx 6 2 |v(x, t)| dx + |u(x, t)| dx , for t > 0,
−∞ −∞ −∞
so that Z ∞
|w(x, t)|2 dx < ∞, for t > 0,
−∞
since both u and v satisfy the integrability conditions in problem (5.321). Sim-
ilarly, we can show that
Z ∞
|wx (x, t)|2 dx < ∞, for t > 0.
−∞
Now, observe that, since both v and u satisfy the initial condition in problem
(5.321,
w(x, 0) = v(x, 0) − u(x, 0) = f (x) − f (x) = 0, for all x ∈ R,
so that w is a solution of problem (5.312) in which the initial condition is the

constant function 0, it follows from the estimate (5.313) in Lemma 5.2.10 that
Z ∞
|w(x, t)|2 dx 6 0, for t > 0,
−∞

Z ∞
|w(x, t)|2 dx = 0, for t > 0. (5.325)
−∞
It follows from (5.325) and the continuity of w that
w(x, t) = 0, for all x ∈ R and t > 0,
so that
v(x, t) = u(x, t), for all x ∈ R and t > 0,
in view of the definition of w in (5.322). Hence, any solution to the problem in
(5.321) must be that given by (5.300).
We will next show that, if u is any solution of problem (5.321), where f

satisfies the integrability condition
Z ∞
|f (x)|2 dx < ∞, (5.326)
−∞
then
lim u(x, t) = 0, for all x ∈ R. (5.327)
t→∞
To see why this is the case, apply Proposition 5.2.11 to write

∞ 2
e−(x−y) /4Dt
Z
u(x, t) = √ f (y) dy,
−∞ 4πDt
for all x ∈ R and t > 0, from which we get that
∞ 2
e−(x−y) /4Dt
Z
|u(x, t)| 6 √ |f (y)| dy, (5.328)
−∞ 4πDt
for all x ∈ R and t > 0. Next, square on both sides of (5.328) and apply the
Cauchy–Schwarz inequality to get
Z ∞ −(x−y)2 /2Dt Z ∞
2 1 e
|u(x, t)| 6 √ √ dy |f (y)|2 dy, (5.329)
8πDt −∞ 2πDt −∞
where 2
∞
e−(x−y) /2Dt
Z
√ dy = 1. (5.330)
−∞ 2πDt
Combining (5.329) and (5.330), we then get
Z ∞
1
|u(x, t)|2 6 √ |f (y)|2 dy, (5.331)
8πDt −∞
lim |u(x, t)|2 = 0, for all x ∈ R,

t→∞
which implies (5.327).
5.3 Solving the Dirichlet Problem in the Unit

Disc
The goal of this section is to construct a solution of the boundary value problem
for the two–dimensional Laplacian

uxx + uyy = 0 in D1 ;
(5.332)
u(x, y) = g(x, y), for (x, y) ∈ ∂D1 ,
where D1 = {(x, y) ∈ R2 | x2 + y 2 < 1} is the unit disk in R2 , and g is a

given function that is continuous in a neighborhood of the unit circle ∂D1 . A
function u ∈ C 2 (D1 , R) that satisfies the PDE in Problem (5.332) is said to be
harmonic. Thus, we would like to find a function, u, that is harmonic in D1
and that takes on the values given by a continuous function, g, on the boundary
of D1 .
To construct a solution of the Dirichlet problem in (5.3) we use the same
procedure that we used to construct a solution of the Vibrating String Problem
in Section 5.1. This will serve as another another illustration of the methods
of separation of variables and eigenfunctions expansion. This approach has
been particularly successful in the construction of solutions of boundary value
problems for linear PDEs over domains with simple geometry.
The discussion here will parallel that of the Vibrating String Problem in
Section 5.1.
• First, in view of the radial symmetry of the domain, we will express prob-
lem (5.3) in polar coordinate r and θ.
5.3. DIRICHLET PROBLEM FOR THE UNIT DISC 141
∂D1
D1
x
Figure 5.3.5: Unit Disk in R2
• Next, we look for a special type of solutions that are products of a function
of r and a function of θ. In other words, we look for solutions in which
the variables separate; this is where we use the method of separation
of variables.
• When looking for solutions that are nonzero over the domain by means of
separation of variables, we are invariable led to an eigenvalue problem.
Solution of the eigenvalue problem leads to a family of solutions in one
(or both of the variables), called eigenfunctions. These eigenfunctions
generate a special family of solutions.
• We will then use the principle of superposition to construct linear combi-

nations of the eigenfunction solutions. We hope that a sequence of these
linear combinations will converge to a function that solves the PDE in
(5.332) and satisfies the boundary condition in that problem; here is where
we use the method of eigenfunctions expansion.
5.3.1 Separation of Variables

In view of the radial symmetry of the domain (see Figure 5.3.5), we will treat
the problem in polar coordinates, (r, θ), where
x = r cos θ and y = r sin θ.
We will also exploit the linearity of the PDE and the boundary condition in
(5.332) and use the principle of superposition to construct a solution of the
problem by superposing simple solutions of the problem. The strategy then is
to, first, find a special class of functions of r and θ that solve Laplace’s equation,
and then use sums of those solutions to construct a solution that also satisfies
the boundary condition.
We begin by expressing the BVP (5.332) in polar coordinates:

 2
∂ v 1 ∂v 1 ∂2v

 + + = 0, 0 < r < 1, −π < θ < π;
∂r2 r ∂r r2 ∂θ2


v(1, θ) = g(cos θ, sin θ), −π 6 θ 6 π,
(5.333)
where we have set
v(r, θ) = u(r cos θ, r sin θ).
We will denote g(cos θ, sin θ) by f (θ), where f : R → R is a continuous, periodic
function of period 2π. We can then rewrite the BVP in (5.333) as
 2
∂ v 1 ∂v 1 ∂2v

 + + = 0, 0 < r < 1, −π < θ < π;
∂r2 r ∂r r2 ∂θ2 (5.334)


v(1, θ) = f (θ), −π 6 θ 6 π.
We start out by looking for special solutions of the PDE in (5.334) of the
form
v(r, θ) = h(r)z(θ), for r > 0 and − π < θ < π, (5.335)
where h : [0, ∞) → R is a continuous functions that is C 2 in (0, ∞), and z : R →
R is a C 2 , periodic function of period 2π. We can therefore compute the partial
derivatives,
∂v
(r, θ) = h0 (r)z(θ), r > 0, −π < θ 6 π;
∂r
∂2v
(r, θ) = h00 (r)z(θ), r > 0, −π < θ 6 π;
∂r2
∂2v
(r, θ) = h(r)z 00 (θ), r > 0, −π < θ 6 π,
∂θ2
and substitute them into the PDE in (5.334) to obtain
1 1
h00 (r)z(θ) + h0 (r)z(θ) + 2 h(r)z 00 (θ) = 0, for r > 0, −π < θ 6 π. (5.336)
r r
Assuming that v(r, θ) is not zero for all values of r and θ, and dividing on both
sides of (5.336) by v(r, θ) as given in (5.335), we obtain
h00 (r) 1 h0 (r) 1 z 00 (θ)

+ + 2 = 0, for 0 < r < 1, −π < θ 6 π. (5.337)
h(r) r h(r) r z(θ)
Multiplying on both sides of the equation in (5.337) by r2 , we notice that that

equation can be written in such a way that the functions that depend only on
r are on one side of the equation and those that depend only on θ are on the
other side of the equation:
h00 (r) h0 (r) z 00 (θ)
r2 +r =− , for 0 < r < 1, −π < θ 6 π. (5.338)
h(r) h(r) z(θ)
Since (5.338) holds true for all values of r and θ in (0, 1) and (−π, π], respectively,
it follows from (5.338) that each side of the equation in (5.338) must be equal
to a constant.2 Call that constant λ so that
h00 (r) h0 (r) z 00 (θ)
r2 +r =− = λ, for 0 < r < 1, −π < θ 6 π. (5.339)
f (r) h(r) z(θ)
The expression in (5.339) leads to two ordinary differential equations
−z 00 (θ) = λz(θ), for − π < θ 6 π, (5.340)
and
r2 h00 (r) + rh0 (r) = λh(r), for r > 0. (5.341)
The requirement that the function g in (5.334) be periodic of period 2π yields
the following conditions for z:
z(−π) = z(π) and z 0 (−π) = z 0 (π); (5.342)
in other words, we will assume that z can be extended to a C 2 periodic function

defined on R with period 2π. Putting together (5.340) and (5.342) yields the
following two–point boundary value problem:
 −z 00 (θ) = λz(θ),

for − π < θ < π;
z(−π) = z(π); (5.343)
 0
z (−π) = z 0 (π).
5.3.2 An Eigenvalue Problem

Observe that the constant function z(θ) = 0, for all values of θ, solves the two–
point BVP in (5.343); we shall refer to this solution as the trivial solution.
We are interested in nontrivial solutions of (5.343); otherwise, the special
solutions in (5.335) of the BVP in (5.334) that we are seeking would all be
the zero function. These solutions will not be helpful in the construction of a
solution of the BVP in (5.334) for arbitrary (nonzero) boundary conditions. We
will see shortly that the answer to the question of whether or not the two–point
BVP in (5.343) has nontrivial solutions depends on the value of λ in the ODE in
that problem. In fact, there is a certain set of values of λ for which (5.343) has
nontrivial solutions; for the rest of the values of λ the two–point BVP (5.343)
has only the trivial solution.
2 To see why this assertion is true, pick θo in (−π, π] such that z(θo ) 6= 0; then, by virtue
h00 (r) h0 (r) z 00 (θo )
of (5.338), r2 +r =− , for all r > 0; so that the left–hand side of (5.338)
h(r) h(r) z(θo )
z 00 (θ)
is constant. Similarly, for fixed ro in (0, 1) with f (ro ) 6= 0, (5.338) implies that =
z(θ)
h00 (r ) h 0 (r )
o o
−ro2 − ro , for all θ in (−π, π], so that the right–hand side of (5.338) must also
h(ro ) h(ro )
be constant.
Definition 5.3.1 (Eigenvalues and Eigenfunctions). A value of λ in (5.343) for

which the two–point BVP in (5.343) has a nontrivial is called an eigenvalue of
the BVP; a corresponding nontrivial solution is called an eigenfunction.
We will next compute the eigenvalues and eigenfunctions of the two–point
BVP in (5.343). Before we proceed with the calculations, it will be helpful to
know that the eigenvalues of (5.343) must be nonnegative. We state that fact
in the following proposition.
Proposition 5.3.2. Assume that the two–point BVP (5.343) has nontrivial
solution. Then, λ > 0.
Proof: Let z be a nontrivial solution of (5.343). Multiply the ODE in (5.343)
by z and integrate from −π to π to get
Z π Z π
− z 00 (θ)z(θ) dθ = λ z(θ)z(θ) dθ. (5.344)
−π −π
Use integration by parts to evaluate the left–most integral in (5.344) to get

Z π π
Z π
z 00 (θ)z(θ) dθ = z(θ)z 0 (θ) − z 0 (θ)z 0 (θ) dθ,
−π −π −π
so that, in view of the boundary conditions in (5.343),

Z π Z π
z 00 (θ)z(θ) dθ = − [z 0 (θ)]2 dθ. (5.345)
−π −π
Substituting the result in (5.345) into the left–hand side of (5.344) then yields
Z π Z π
[z 0 (θ)]2 dθ = λ [z(θ)]2 dθ. (5.346)
−π −π
Since
Z π z is a nontrivial solution of the two–point BVP in (5.343), it follows that
[z(θ)]2 dθ > 0. We can therefore solve (5.346) for λ to obtain
−π
Z π
[z 0 (θ)]2 dθ
λ = Z−π
π ,
[z(θ)]2 dθ
−π
which shows that λ is nonnegative.

In view of the result of Proposition 5.3.2, it suffices to look for nontrivial
solutions of (5.343) for either λ = 0 or λ > 0.
For the case in which λ = 0 in (5.343), the ODE in (5.343) becomes
z 00 (θ) = 0,
which has general solution

z(θ) = c1 θ + c2 , (5.347)
for arbitrary constants c1 and c2 .
Applying the first boundary condition to z given in (5.347) yields
−πc1 + c2 = πc2 + c2 ,
from which we get that 2πc1 = 0, so that c1 = 0. It then follows from (5.347)
any solution of the BVP in (5.343) with λ = 0 must be constant:
z(θ) = c, for all θ. (5.348)
In particular, if c 6= 0 in (5.348), z(θ) = c for all θ is a nontrivial solution of the

two–point BVP (5.343). Consequently, λ = 0 is an eigenvalue of (5.343). For
future reference, we shall denote this eigenvalue by λo , so that
λo = 0, (5.349)
and we shall pick the special eigenfunction
ϕo (θ) = 1, for all θ, (5.350)
and note that any solution of the BVP in (5.343) for λo is a constant multiple
of ϕo given in (5.350); so that
zo (θ) = ao , for all θ, (5.351)
where ao denotes a real constant, represents all solutions of the two–point BVP
in (5.350) corresponding to the eigenvalue λo = 0.
Next, we look for positive eigenvalues of the BVP in (5.343). For the case
in which λ > 0 in (5.343), the general solution of the ODE in (5.343) is
√ √
z(θ) = c1 cos( λθ) + c2 sin( λθ), for all θ, (5.352)
and arbitrary constants c1 and c2 , so that

√ √ √ √
z 0 (θ) = −c1 λ sin( λθ) + c2 λ cos( λθ), for all θ. (5.353)
Imposing the the boundary conditions in (5.343) to the functions given in (5.352)
and (5.353) yields the system of equations
√ √ √ √
√ c1 cos(−√ λπ) +√c2 sin(−√λπ) = c1 cos( √ λπ)√+ c2 sin( √λπ); √
−c1 λ sin(− λπ) + c2 λ cos(− λπ) = −c1 λ sin( λπ) + c2 λ cos( λπ);
√ (5.354)
thus, dividing the second equation in (5.354) by λ, since λ > 0, and using the
fact that cos is even and sin is odd,
√ √ √ √
c1 cos( √λπ) − c2 sin(−√λπ) = c1 cos( √ λπ) + c2 sin( √λπ);
c1 sin( λπ) + c2 cos( λπ) = −c1 sin( λπ) + c2 cos( λπ),

√
2c2 sin(√λπ) = 0;
(5.355)
2c1 sin( λπ) = 0.
Since we are looking for nontrivial solutions of (5.343), we require that c1 and
c2 in (5.352) are not both zero. Consequently, we obtain from (5.355) that
√
sin( λπ) = 0. (5.356)
Solutions of the trigonometric equation in (5.356) are given by
√
λπ = nπ (5.357)
where n is an integer. It follows from (5.357) that the positive eigenvalues of
the BVP in (5.343) are given by
λ = n2 , for n = 1, 2, 3, . . . . (5.358)
We will denote the positive eigenvalues of the BVP (5.343) in (5.358) by λn , for
n = 1, 2, 3, . . ., so that
λn = n2 , for n = 1, 2, 3, . . . . (5.359)
We will denote the corresponding eigenfunctions by zn . These are linear com-
binations of cos(nθ) and sin(nθ), so that
zn (θ) = an cos(nθ) + bn sin(nθ) for n = 1, 2, 3, . . . , and θ ∈ R, (5.360)
where an and bn , for n = 1, 2, 3 . . ., are real constants.
We shall put together the results in (5.349), (5.351), (5.359) and (5.360) in
the following proposition:
Proposition 5.3.3 (Eigenvalues and eigenfunctions of BVP (5.343)). The eigen-
values of the two–point BVP (5.343) are given by
λ n = n2 , for n = 0, 1, 2, 3, . . . , (5.361)
with corresponding eigenfunctions of the form
ϕo (θ) = 1, for all θ ∈ R,
corresponding to λo = 0, and
ϕn1 (θ) = cos(nθ) and ϕn2 (θ) = sin(nθ) for θ ∈ R and n = 1, 2, 3, . . . ,
2
corresponding to λn = n , for n = 1, 2, 3, . . ..
The corresponding eigenspaces are made up of the funnctions
zo (θ) = ao , for all θ ∈ R,
and arbitrary real numbers ao , and
zn (θ) = an cos(nθ) + bn sin(nθ) for θ ∈ R and n = 1, 2, 3, . . . ,
where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are arbitrary real
constants.
With the values for λ given in (5.361), we now proceed to solve the ODE in
(5.341) for the radial component of the special solutions of the BVP in (5.334)
of the form given in (5.335); namely,
r2 h00 (r) + rh0 (r) = n2 h(r), for r > 0 and n = 0, 1, 2, . . . (5.362)
We shall first solve (5.362) for the case n = 0. In this case the equation becomes
rh00 (r) + h0 (r) = 0, for r > 0, (5.363)
where we have divided by r > 0. Observe that the equation in (5.363) can be
written as
d
[rh0 (r)] = 0, for r > 0,
dr
which can be integrated to yield
rh0 (r) = c1 , for r > 0,
and some constant c1 , or

c1
h0 (r) = , for r > 0, (5.364)
r
and some constant c1 . Integrating the equation in (5.364) then yields
h(r) = c1 ln(r) + c2 , for r > 0, (5.365)
and some constants c1 and c2 . Observe that, if c1 6= 0 in (5.365), the function

h given a (5.365) is unbounded as r → 0+ . Thus, since we are looking for
C 2 functions defined in the unit disc, D1 , we must set c1 equal to 0. This is
equivalent to imposing the following boundary condition on h:
lim h(r) exists. (5.366)

r→0+
Hence, it follows from (5.365) and (5.366) that, for n = 0, a solution of (5.362)
is given by
h(r) = c, for all r, (5.367)
is a solution, for some constant c. Taking c = 1 in (5.367) we get the solution
of (5.362) corresponding to n = 0:
ho (r) = 1, for all r. (5.368)
Next, consider the case n > 1 in (5.362). In this case the differential equation
in (5.362) is an ODE of Euler type:
r2 h00 (r) + rh0 (r) − n2 h(r) = 0, for r > 0. (5.369)
The ODE in (5.369) can be solved by looking for solutions of the form
h(r) = rq , for r > 0, (5.370)

and some real number q.

Taking derivatives of h in (5.370) and substituting into (5.370) yields
r2 q(q − 1)rq−2 + rqrq−1 − n2 rq = 0, for r > 0,
or
q(q − 1)rq + qrq − n2 rq = 0, for r > 0,
or
[q(q − 1) + q − n2 ]rq = 0, for r > 0. (5.371)
q(q − 1) + q − n2 = 0,
or
q 2 − n2 = 0,
or
(q + n)(q − n) = 0,
q = ±n. (5.372)
h−n (r) = r−n and hn (r) = rn , for r > 0. (5.373)
In view of the boundary condition in (5.366), we take the second solution in

(5.373),
hn (r) = rn , for all r and n = 1, 2, 3, . . . . (5.374)
Putting together (5.374), (5.368), (5.360), (5.351), and (5.335), we conclude
that we have found an infinite collection of solutions of the PDE in (5.334);
namely,
vo (r, θ) = ao , for all r and θ; (5.375)
vn (r, θ) = rn [an cos(nθ) + bn sin(nθ)], for all r and θ, (5.376)
where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are real constants.
5.3.3 Expansion in Terms of Eigenfunctions

None of the functions in (5.375) and (5.376) by itself will satisfy the general
boundary condition in (5.334). We can, however, attempt to construct a solution
of (5.334) by adding all of them together; in other words, by applying the
principle of superposition:
∞
X
v(r, θ) = ao + rn [an cos(nθ) + bn sin(nθ)], 0 6 r < 1, −π < θ 6 π. (5.377)
n=0
provided the series in (5.377) converges to a C 2 function.

Let’s assume for the moment that the series in (5.377) converges also for
r = 1, so that we can apply the boundary condition in (5.334) to get
∞
X
ao + [an cos(nθ) + bn sin(nθ)] = f (θ), for − π < θ 6 π. (5.378)
n=0
Assuming for the moment that the series on the left–hand side of (5.378) con-
verges in such a way that it can be integrated term by term, we can compute
the values of the coefficients an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., in
terms of the function f by means of the following integration facts:
Z π
sin(nθ) cos(mθ) dθ = 0, for all m, n = 1, 2, 3, . . . ; (5.379)
−π
(
π
if m 6= n;
Z
0,
cos(nθ) cos(mθ) dθ = (5.380)
−π π, if m = n;
and (
π
if m 6= n;
Z
0,
sin(nθ) sin(mθ) dθ = (5.381)
−π π, if m = n.
Indeed, integrating on both sides of (5.378) from −π to π we get, assuming that

the series in (5.378) can be integrated term by term,
Z π
2πao = f (θ) dθ,
−π
from which we get Z π

1
ao = f (θ) dθ; (5.382)
2π −π
thus, ao is the average value of f over the interval (−π, π].

Next, multiply the equation in (5.378) on both sides by cos(mθ) to obtain
∞
X
ao cos mθ + [an cos nθ cos mθ + bn sin nθ cos mθ] = f (θ) cos mθ. (5.383)
n=0
Then, integrate on both sides of (5.383) with respect to θ from −π to π, and

use the identities in (5.379) and (5.380) to get
Z π
πam = f (θ) cos(mθ) dθ,
−π
from which we get

Z π
1
am = f (θ) cos(mθ) dθ, for m = 1, 2, 3, . . . . (5.384)
π −π
Similar calculations (this time multiplying the equation in (5.378) on both sides
by sin(mθ), integrating from −π to π, and using the integral identities in (5.379)
and (5.381)) lead to
1 π
Z
bm = f (θ) sin(mθ) dθ, for m = 1, 2, 3, . . . . (5.385)
π −π
The numbers defined in (5.382), (5.384) and (5.385) are the Fourier coef-
ficients of the 2π–periodic function f . Note that the Fourier coefficients of f
are defined whenever f is absolutely integrable over the interval [−π, π].
Definition 5.3.4 (Absolute Integrability). A function f : [−π, π] → R is said
to be absolutely integrable over [−π, π] whenever
Z π
|f (θ)| dθ < ∞. (5.386)
−π
Note that f doesn’t have to be continuous for (5.386) to hold true. For
instance, if f is bounded and piecewise continuous then (5.386) holds true;
indeed, suppose that f piecewise continuous and
|f (θ)| 6 M, for θ ∈ [−π, π],
and some positive constant M ; then

Z π Z π
|f (θ)| dθ 6 M dθ = 2πM < ∞.
−π −π
Notation 5.3.5. We will denote the integral in (5.386) by kf kL1 ; so that

Z π
kf kL =
1 |f (θ)| dθ. (5.387)
−π
If the integral in (5.387) ia understood as the Lebesgue integral, and kf kL1 < ∞,
we will say that f is an L1 function and write f ∈ L1 (−π, π). We shall refer to
kf kL1 as the L1 norm of f ∈ L1 (−π, π).
The existence of the Fourier coefficients of f in (5.382), (5.384) and (5.385)
is guaranteed for absolutely integrable 2π–periodic functions, f , or for f ∈
L1 (−π, π). This is the content of the following proposition.
Proposition 5.3.6 (Existence of the Fourier Coefficients). Let an , for n =
0, 1, 2, . . ., be as given in (5.382) and (5.384), and bn , for n = 1, 2, 3, . . ., be as
in (5.385), where f ∈ L1 (−π, π). Then,
1
|an | 6 kf kL1 , for n = 0, 1, 2, 3, . . . ; (5.388)
π
and
1
|bn | 6 kf kL1 , for n = 1, 2, 3, . . . . (5.389)
π
Proof: The estimates in (5.388) and (5.389) follow from properties of the inte-
gral. For ao , we get from (5.382) that
Z π
1
|ao | 6 |f (θ)| dθ,
2π −π
so that, using the definition of the L1 norm of f in (5.387),
1 1
|ao | 6 kf kL1 6 kf kL1 .
2π π
For n = 1, 2, 3, . . . we obtain from (5.384) that
1 π
Z
|an | 6 |f (θ)| | cos(nθ)| dθ
π −π
Z π
1
6 |f (θ)| dθ,
π −π
since | cos(nθ)| 6 1 for all θ and all n, which yields (5.388). Similar calculations
lead to (5.389).
It follows from Proposition 5.3.6 that the sequences of Fourier coefficients,

(an ) and (bn ), of f are bounded by a constant depending on the L1 norm of
f . In fact, it can be shown that the Fourier coefficients of an L1 , 2π–periodic
functions tend to 0 as n goes to infinity; this is known as the Riemann–Lebesgue
Lemma.
Proposition 5.3.7 (Riemann–Lebesgue Lemma). Let an , for n = 0, 1, 2, . . .,

be as given in (5.382) and (5.384), and bn , for n = 1, 2, 3, . . ., be as in (5.385),
where f ∈ L1 (−π, π). Then,
lim an = 0 and lim bn = 0.

n→∞ n→∞
For a proof of the Riemann-Lebesgue Lemma, see [Tol62].

We will next use the result of Proposition 5.3.6 to prove that the series
defining the function v in (5.377) converges in D1 .
Proposition 5.3.8 (Point–wise Convergence of Series in (5.377) ). Let an , for

n = 0, 1, 2, . . ., be as given in (5.382) and (5.384), and bn , for n = 1, 2, 3, . . .
be as in (5.385), where f ∈ L1 (−π, π). Then, the series defining v in (5.377)
converges absolutely in D1 .
Proof: The conclusion will follow by comparing with the geometric series since
0 6 r < 1 and
kf kL1 n
|rn an cos(nθ)| 6 rn |an | 6 r , for all n,
π
where we have used the estimate (5.388) in Proposition 5.3.6. Similarly, using
(5.389) in Proposition 5.3.6,
kf kL1 n
|rn bn sin(nθ)| 6 r ,
π
for all n.
Proposition 5.3.8 allows us to conclude that the function v given in (5.377)

is well defined. However, in order to prove that that function is harmonic in
D1 , we have to be able to differentiate the series term by term. This would
be possible, for instance, if we knew that the series on the right–hand–side of
(5.377), and the series for the partial derivatives
∞
X
nrn [−an sin(nθ) + bn cos(nθ)],
n=0
∞
X
− n2 rn [an cos(nθ) + bn sin(nθ)],
n=0
∞
X
nrn−1 [an cos(nθ) + bn sin(nθ)],
n=0
and
∞
X
n(n − 1)rn−2 [an cos(nθ) + bn sin(nθ)],
n=0
converge uniformly. However, we do not know that at this point. In order to

answer these questions, though, we will have to make further assumptions on f .
Before we deal with these questions, we will first answer the question of when the
trigonometric series on the left–hand side of the equation in (5.378) converges
uniformly. Uniform convergence will justify the term–by–term integration that
was done in order to obtain the formulas in (5.382), (5.384) and (5.385). We will
denote the trigonometric series on the left–hand side of the equation in (5.378)
by fb(θ), so that
∞
X
fb(θ) = ao + [an cos(nθ) + bn sin(nθ)], for − π 6 θ 6 π, (5.390)
n=0
where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are the Fourier coef-

ficients of f . Let (fbn (θ)) denote the sequence of partial sums of the series in
(5.390) so that
n
X
fbn (θ) = ao + [ak cos(kθ) + bk sin(kθ)], for − π 6 θ 6 π. (5.391)
k=0
Definition 5.3.9 (Uniform Convergence). We say that sequence of functions,

(fbn ), defined in (5.391) converges uniformly to f in [−π, π] if
lim max |fbn (θ) − f (θ)| = 0.

n→∞ −π6θ6π
The following proposition gives a sufficient condition for the trigonometric

series in (5.390) to converge uniformly to f .
Theorem 5.3.10 (Uniform Convergence 3). Assume that f : R → R is a con-
tinuous, 2π–periodic function; assume also that f is piecewise differentiable with
f 0 : R → R piecewise continuous. Let (fbn ) be the sequence of trigonometric func-
tions defined in (5.390), where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . .,
are the Fourier coefficients of f . Then, (fbn ) converges uniformly to g in [−π, π]
as n → ∞.
The proof of Theorem 5.3.10 follows from the Uniform Convergence Theorem
2 in Theorem 5.1.14, with L = π, on page 111 in these notes. Note that, if f is
piecewise differentiable with piecewise continuous derivative, f 0 : R → R, then
f 0 is square–integrable on [−π, π]. (For another proof, see [Tol62, pp. 80-81]).
Let’s assume for the moment that f : R → R is 2π–periodic, and satisfies
the assumptions of 5.3.10. It then follows from Theorem 5.3.10 that the Fourier
series on the left–hand side of (5.378), where an , for n = 0, 1, 2, . . ., and bn ,
for n = 1, 2, 3, . . ., are the Fourier coefficients of f , converges uniformly to the
right–hand side of the equation. This justifies the term–by–term integration of
the series that lead to the formulas for the Fourier coefficients in (5.382), (5.384)
and (5.385) by virtue of the following theorem form Analysis:
Theorem 5.3.11 (Term–by–Term Integration). Let (uk ) be a sequence of con-
tinuous functions over a closed and bounded interval, [a, b]. Assume that the
series
∞
X
uk
k=1
converges uniformly to f . Then, f is continuous on [a, b], and

Z b ∞ Z b
X
f (x) dx = uk (x) dx,
a k=1 a
or
∞ ∞ Z
!
Z b X X b
uk (x) dx = uk (x) dx,
a k=1 k=1 a
For a proof of this theorem refer to Rudin [Rud53, pg. 121–122].

We saw in Proposition 5.3.8 that the trigonometric series defining v(r, θ) in
(5.377),
∞
X
v(r, θ) = ao + rn [an cos nθ + bn sin nθ], 0 6 r 6 1, −π 6 θ 6 π, (5.392)
n=0
converges in D1 , provided that f ∈ L1 (−π, π), or absolutely integrable on

[−π, π]. In the next proposition we will use the Weierstrass M–Test for uni-
form convergence, (see [Rud53, Theorem 7.10, pg. 119]), in order to show that,
for the case in which f is piecewise C 1 and 2π–periodic, then the series in
(5.392), where the an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are the
Fourier coefficients of f , converges uniformly in D1 , the closed unit disk in R2 .
Proposition 5.3.12 (Uniform Convergence of Series in (5.377) ). Let an , for

n = 0, 1, 2, . . ., be as given in (5.382) and (5.384), and bn , for n = 1, 2, 3, . . . be
as in (5.385), where f is a piecewise C 1 , 2π–periodic function. Then, the series
defining v in (5.392) converges uniformly in D1 .
Proof: The assumptions that f is piecewise C 1 and 2π–periodic imply that the
Fourier coefficients of f satisfy the estimate
∞
X
(|ak | + |bk |) < ∞. (5.393)
k=0
See the calculations leading to the proof of Proposition 5.1.9, with L = π, on

page 106 in these notes for a derivation of the estimate in (5.393).
Next, use the triangle inequality to estimate the absolute values of the terms
of the series in (5.392) to get
|rn [an cos(nθ) + bn sin(nθ)]| 6 |an | + |bn |, for all n = 1, 2, 3, . . . ,
and all r ∈ [0, 1] and θ ∈ [−π, π]. Thus, the absolute values of the terms of
the series in (5.392) are “majorized” by the terms of the convergent series in
(5.393). It then follows by the the Weierstrass M–Test for uniform convergence
([Rud53, Theorem 7.10, pg. 119]) that the series in (5.392) converges uniformly
for r ∈ [0, 1] and θ ∈ [−π, π].
We will get a chance to use the Weierstrass M–Test for uniform convergence
once again to justify the following calculations based on the trigonometric series
representation for v(r, θ) in (5.392) and the assumption that f is a piecewise
C 1 , 2π–periodic function.
First, substitute the formulas defining the Fourier coefficients of f in (5.382),
(5.384) and (5.385) into the right–hand side (5.392) to get

Z π ∞ Z π
1 X 1
v(r, θ) = f (ξ) dξ + rn f (ξ) cos(nξ) dξ cos(nθ)
2π −π n=0
π −π
Z π
1
+ f (ξ) sin(nξ) dξ sin(nθ)
π −π
Z π ∞ Z π
1 1X n
= f (ξ) dξ + r cos(nθ) cos(nξ)f (ξ) dξ
2π −π π n=0 −π
Z π
+ sin(nθ) sin(nξ)f (ξ) dξ
−π
Z π
1
= f (ξ) dξ
2π −π
∞ Z
1X π n
+ r [cos(nθ) cos(nξ) + sin(nθ) sin(nξ)]f (ξ) dξ,
π n=0 −π

Z π ∞ Z
1 1X π n
v(r, θ) = f (ξ) dξ + r cos[n(θ − ξ)]f (ξ) dξ, (5.394)
2π −π π n=0 −π
for 0 6 r < 1 and θ ∈ [−π, π], by virtue of the trigonometric identity
cos(α − β) = cos α cos β + sin α sin β.
Next, we will interchange the order of integration and summation in (5.394).

This is justified by the fact that the series
∞
X
rn cos(nξ)
n=1
converges absolutely and uniformly in ξ ∈ [−π, π] for 0 6 r < 1. To see why

this is the case, note that
|rn cos(nξ)| 6 rn , for all n = 1, 2, 3, . . .
Thus, the assertion follows by the Weierstrass M–Test for uniform convergence,
for 0 6 r < 1.
Hence, interchanging the order of summation and integration in (5.394), we
can write
Z π " ∞
#
1 X
v(r, θ) = 1+ 2rn cos[n(θ − ξ)] f (ξ) dξ, (5.395)
2π −π n=1
for 0 6 r < 1 and θ ∈ [−π, π]. Putting

∞
" #
1 X
n
P (r, θ) = 1+ 2r cos(nθ) , for 0 6 r < 1 and θ ∈ [−π, π], (5.396)
2π n=1
we see that (5.397) can be written as

Z π
v(r, θ) = P (r, θ − ξ)f (ξ) dξ, for 0 6 r < 1 and θ ∈ [−π, π]. (5.397)
−π
5.3.4 The Poisson Kernel for the Unit Disc

The function P defined in (5.396) is called the Poisson kernel for the unit
disc in R2 , and the expression on the right–hand side of (5.397) is called the
Poisson integral representation for v. In this section and the next, we will
prove several important properties of the Poisson kernel and the Poisson integral
in (5.397).
We will first show that the series defining the Poisson kernel in (5.396)
converges uniformly over θ ∈ [−π, π] for each 0 6 r < 1. This will justify
term–by–term integration of the series. We will use the Weierstrass M–Test for
uniform convergence. Thus, we first estimate the absolute values of the of the
terms of the series,
|2rn cos(nθ)| 6 2rn , for all θ ∈ [−π, π], (5.398)
and for n ∈ N. It follows from (5.398) that the absolute values of the terms of
the series in (5.396) are “majorized” by the terms of the convergent geometric
series
X∞
2rn ,
n=1
for 0 6 r < 1. Hence, the Weierstrass M–Test applies, and we conclude that
the series defining P (r, θ) in (5.396) converges uniformly in θ for 0 6 r < 1.
This arguments can be carried out further to prove that, for any 0 < R < 1,
the series defining P (r, θ) in (5.396) converges uniformly for θ ∈ [−π, π] and
r ∈ [0, R]. Hence, P (r, θ) defines a continuous function in the open unit disc,
D1 , in R2 . This follows from the following important consequence of the uniform
convergence of a sequence of continuous functions:
Proposition 5.3.13 (Uniform Limit of Continuous Functions). Let (fn ) be a

sequence of continuous functions on [a, b] that converges uniformly to a function
f : [a, b] → R. Then, f is continuous.
For a proof of this proposition see Rudin [Rud53, Theorem 7.12, pg. 20].
We will next show that, in fact, the Poisson kernel is C 2 in D1 and that it
solves Laplace’s equation in D1 . In order to show that the partial derivatives of
P exist, we need to show that the series

∞
X ∞
X
2nrn−1 cos(nθ) and 2nrn sin(nθ) (5.399)
n=1 n=1
converge uniformly. This assertion will follow from the following proposition
Proposition 5.3.14 (Term–by–Term Differentiation). Let (uk ) be a sequence
of functions that are differentiable over a closed and bounded interval, [a, b].
Assume that the series
X∞
u0k
k=1
converges absolutely and uniformly over [a, b]. Assume also that the series
∞
X
uk (xo )
k=1
converges absolutely at some point xo in [a, b]. Then, the series converges
∞
X
uk
k=1
converges uniformly to a function f that is differentiable over [a, b], and

∞
X
f 0 (x) = u0k (x), for all x ∈ [a, b];
k=1
or "∞ ∞
#
d X X
uk (x) = u0k (x) ,
dx
k=1 k=1
for all x ∈ [a, b].

This proposition can be proved by applying Theorem 7.17 in [Rud53, pg.
124].
In order to see that the series in (5.399) converge absolutely and uniformly,
first note that
|2nrn−1 cos(nθ)| 6 2nrn−1 and |2nrn sin(nθ)| 6 2nr2
for all θ ∈ [−π, π]; so that the series in (5.399) are “majorized” by the series
∞
X ∞
X
2nrn−1 and 2nrn , (5.400)
n=1 n=1
respectively; both of the series in (5.400) converge by the Ratio Test (or the
Root Test), since 0 6 r < 1. It then follows by the Weierstrass M–Test for
uniform convergence that the series in (5.400) converge uniformly in θ. The

same argument applied to r ∈ [0, R], where where R < 1, yields that the series
in (5.399) are absolutely and uniformly convergent for θ ∈ [−π, π] and r ∈ [0, R].
This time the series in (5.399) are “majorized” by the convergent series
∞
X ∞
X
2nRn−1 and 2nRn ,
n=1 n=1
respectively. It then follows from Proposition 5.3.14 that the partial derivatives
of the Poisson kernel in (5.396) have partial derivatives in D1 given by
∞
∂ 1 X
[P (r, θ)] = 2nrn−1 cos(nθ), 0 6 r < 1 and θ ∈ [−π, π], (5.401)
∂r 2π n=1
and
∞
∂ 1 X
[P (r, θ)] = − 2nrn sin(nθ), 0 6 r < 1 and θ ∈ [−π, π], (5.402)
∂θ 2π n=1
where we have differentiated the series in (5.396) term–by–term. A similar

argument can be used to obtain the second partial derivatives of the Poisson
kernel:
∞
∂2 1 X
[P (r, θ)] = 2n(n − 1)rn−2 cos(nθ), 0 6 r < 1, θ ∈ [−π, π], (5.403)
∂r2 2π n=1
and
∞
∂2 1 X 2 n
[P (r, θ)] = − 2n r cos(nθ), 0 6 r < 1 and θ ∈ [−π, π], (5.404)
∂θ2 2π n=1
where the series in (5.401) and (5.402) have been differentiated term–by–term.
Next, substitute the partial derivatives in (5.401), (5.403) and (5.404) into
the expression for the Laplacian of P in polar coordinates to get
∞
∂2P 1 ∂P 1 ∂2P 1 X
+ + = 2n(n − 1)rn−2 cos(nθ)
∂r2 r ∂r r2 ∂θ2 2π n=1
∞
1 X
+ 2nrn−1 cos(nθ)
2πr n=1
∞
1 X 2 n
− 2n r cos(nθ) (5.405)
2πr2 n=1
∞
1X
= [n(n − 1) + n − n2 ]rn−2 cos(nθ)
π n=1
= 0,
for all θ ∈ [−π, π] and 0 6 r < 1. We have therefore shown that the Poisson
kernel solves Laplace’s equation in the open unit disc.
Next, integrate the series in (5.396) over the interval [−π, π], which is justi-
fied by the uniform convergence of the series, to obtain
Z π
P (r, θ) dθ = 1, for all 0 6 r < 1. (5.406)
−π
The series defining the Poisson kernel in (5.396) can actually be evaluated
by using the identity
2 cos(nθ) = einθ + e−inθ ,
and then adding geometric series. Indeed,
∞
X ∞
X
2rn cos(nθ) = rn [einθ + e−inθ ]
n=1 n=1
∞
X ∞
X
= rn [eiθ ]n + rn [e−iθ ]n
n=1 n=1
∞
X ∞
X
= [reiθ ]n + [re−iθ ]n ,
n=1 n=1
so that, since |re±iθ | = r < 1, for all θ,

∞
X reiθ re−iθ
2rn cos(nθ) = + ,
n=1
1 − re iθ 1 − re−iθ
which simplifies to
∞
X reiθ − r2 + re−iθ − r2
2rn cos(nθ) = ,
n=1
1 − reiθ − re−iθ + r2
or
∞
X r[eiθ + e−iθ ] − 2r2
2rn cos(nθ) = ,
n=1
1 − r[eiθ + e−iθ ] + r2
or
∞
X 2r cos(θ) − 2r2
2rn cos(nθ) = , 0 6 r < 1, θ ∈ [−π, π]. (5.407)
n=1
1 − 2r cos(θ) + r2
Substituting the value of the series in (5.407) into (5.396) then yields the formula
1 1 − r2
P (r, θ) = , for 0 6 r < 1 and θ ∈ [−π, π], (5.408)
2π 1 − 2r cos(θ) + r2
for the Poisson kernel.
We will next use the formula in (5.408) for the Poisson kernel for the unit disk
to derive further properties of the Poisson kernel. We summarize these proper-
ties, as well as the ones we have already established using the representation in
(5.396), in the following proposition.
Proposition 5.3.15 (Properties of the Poisson Kernel). Let P (r, θ) be given

by (5.408), or its equivalent representation as an infinite series in (5.396). Then,
the function P : [0, 1) × [−π, π] → R satisfies the following:
(i) P (r, θ) > 0 for all (r, θ) ∈ [0, 1) × [−π, π];
(ii) P ∈ C ∞ ([0, 1) × [−π, π]);
(iii) P is harmonic in D1 ;
Z π
(iv) P (r, θ − ξ) dξ = 1, for all ξ ∈ R and all 0 6 r < 1.
−π
(v) lim− P (r, θ − ξ) = 0, for ξ 6= θ and |ξ − θ| < π;

r→1
(vi) lim− P (r, θ − ξ) = +∞, for ξ = θ.

r→1
Proof: In order to prove (i) and (ii), first note that, for all θ ∈ R and r > 0,
2r cos θ 6 2r,
so that
1 − 2r cos θ + r2 > 1 − 2r + r2 ,
or
1 − 2r cos(θ) + r2 > (1 − r)2 , for 0 6 r < 1 and θ ∈ R. (5.409)
It follows from (5.409) and the formula for P (r, θ) in (5.408) that P (r, θ) is
defined for all r ∈ [0, r) and all θ ∈ R, and P (r, θ) > 0 for r ∈ [0, 1) and all
θ ∈ R; we have therefore establlished (i).
From (5.409) we also obtain that
1 − 2r cos(θ) + r2 > 0, for 0 6 r < 1 and θ ∈ R.
Thus, the denominator in the formula for P (r, θ) in (5.408) is not zero for
0 6 r < 1 and θ ∈ R; hence, since the numerator and denominator of the
expression defining P (r, θ) in (5.408) are C ∞ functions, (ii) also follows.
We have already established that P satisfies Laplace’s equation in D1 (see
the calculations leading up to (5.405) on page 158) using the definition of P in
(5.396). Thus, P is harmonic in D1 and so we have established (iii).
The integral identity in (iv) will follow from (5.406) and the 2π–periodicity
of P (r, θ) in θ. Indeed, making the change of variables ζ = θ − ξ in the integral
in (iv) we have
Z π Z θ−π
P (r, θ − ξ) dξ = − P (r, ζ) dζ
−π θ+π
Z θ+π
= P (r, ζ) dζ
θ−π
Z π
= P (r, ζ) dζ
−π
= 1,
for all θ ∈ R.
Next, use the formula for P (r, θ) in (5.408) to obtain that
1 1 − r2
P (r, θ − ξ) = ,
2π 1 − 2r + r2
for θ = ξ, from which we get that
1 1+r
P (r, θ − ξ) = , for 0 6 r < 1 and θ = ξ. (5.410)
2π 1 − r
The assertion in (vi) follows from (5.410).
To prove (v), first note that
lim [1 − 2r cos(θ − ξ) + r2 ] = 2 − 2 cos(θ − ξ)

r→1−
= sin2 (θ − ξ)
so that
lim [1 − 2r cos(θ − ξ) + r2 ] 6= 0, for ξ 6= 0 and |ξ − θ| < π (5.411)

r→1−
The assertion in (v) then follows from (5.411) and the expression for the Poisson
kernel in (5.408).
5.3.5 The Poisson Integral Representation

Let f : [−π, π] → R be a continuous function that can be extended to a contin-
uous, 2π–periodic function in R. We then have that
|f (θ)| 6 M, for all θ ∈ [−π, π],
and some positive constant M . The goal of this section is to use the properties
of the Poisson kernel listed in Proposition 5.3.15 to prove that the function
v : D1 → R defined by
Z π


 P (r, θ − ξ)f (ξ) dξ, for 0 6 r < 1, θ ∈ [−π, π];
−π
v(r, θ) = (5.412)



f (θ), for r = 1, θ ∈ [−π, π],
where P (r, θ) denotes the Poisson kernel for the unit disc in R2 given in (5.396)
or (5.408), solves the Dirichlet problem (5.334) for the unit disc in R2 .
We first show that v ∈ C 2 (D1 , R) and that it solves Laplace’s equation in
D1 ; in polar coordinates,
∂ 2 v 1 ∂v 1 ∂2v
+ + = 0, for 0 < r < 1, θ ∈ [−π, π]. (5.413)
∂r2 r ∂r r2 ∂θ2
This will follow from (iii) in Proposition 5.3.15, provided we can show that
differentiation under the integral sign in the first part of the definition of v in
(5.412) is valid. Indeed, property (iii) in Proposition 5.413 says that
∂2P 1 ∂P 1 ∂2P
2
+ + 2 = 0, for 0 6 r < 1, θ ∈ [−π, π]. (5.414)
∂r r ∂r r ∂θ2
Thus, assuming for the moment that differentiation under the integral sign in
(5.412) is valid, we have that, for 0 6 r < 1 and θ ∈ [−π, π],
Z π 2
∂ 2 v 1 ∂v 1 ∂2v ∂
2
+ + 2 2 = 2
[P (r, θ − ξ)]f (ξ) dξ
∂r r ∂r r ∂θ −π ∂r
Z π
1 ∂
+ [P (r, θ − ξ)]f (ξ) dξ
−π r ∂r
π
1 ∂2
Z
+ [P (r, θ − ξ)]f (ξ) dξ,
−π r2 ∂θ2
Z π 2
1 ∂2

∂ 1 ∂
∆v = 2
[P (r, θ − ξ)] + [P (r, θ − ξ)] + 2 2 [P (r, θ − ξ)] f (ξ) dξ,
−π ∂r r ∂r r ∂θ
where we have used the short–hand notation, ∆v, for the Laplacian of v. The
fact that v is harmonic in D1 then follows from the previous identity and (5.414).
We will next see that differentiation under the integral sign is justified. In
order to do this, we first note that the continuity of f implies that there exists
a positive constant, M , such that
|f (θ)| 6 M, for all θ ∈ [−π, π]. (5.415)
In view of (5.415) and (5.414), in order to justify the differentiation under the
integral sign in the first part of the definition of u in (5.412), it suffices to prove
that
∂ ∂ ∂2
[P (r, θ − ξ)], [P (r, θ − ξ)] and [P (r, θ − ξ)]
∂θ ∂r ∂θ2
are absolutely integrable over [−π, π] for each [−π, π].

Use (5.408) to compute
∂ 1 (1 − r2 )2r sin(θ − ξ)
[P (r, θ − ξ)] = − ,
∂θ 2π (1 − 2r cos(θ − ξ) + r2 )2

∂ 2r sin(θ − ξ)
[P (r, θ − ξ)] = − P (r, θ − ξ), (5.416)
∂θ 1 − 2r cos(θ − ξ) + r2
by virtue of the expression for the Poisson kernel in (5.408). Next, take absolute
values on both sides of (5.416) and use the estimate in (5.409) to get
∂ 2r
[P (r, θ − ξ)] 6 P (r, θ − ξ), (5.417)
∂θ (1 − r)2
where we have used the positivity of the Poisson kernel in (i) of Proposition
5.3.15. Integrating on both sides of the inequality in (5.417) form −π to π and
using property (iv) in Proposition 5.3.15 we obtain that
Z π
∂ 2r
[P (r, θ − ξ)] dξ 6 , for 0 6 r < 1 and θ ∈ [−π, π],
−π ∂θ (1 − r)2
∂
which shows that [P (r, θ − ξ)] is absolutely integrable over [−π, π] for 0 6
∂θ
r < 1 and θ ∈ [−π, π].
Next, take partial derivative with respect to θ on both sides of (5.416) to get
∂2 2r cos(θ − ξ)
[P (r, θ − ξ)] = − P (r, θ − ξ)
∂θ2 1 − 2r cos(θ − ξ) + r2
4r2 sin2 (θ − ξ)
+ P (r, θ − ξ)
(1 − 2r cos(θ − ξ) + r2 )2
2r sin(θ − ξ) ∂
− 2
[P (r, θ − ξ)],
1 − 2r cos(θ − ξ) + r ∂θ
∂2 2r cos(θ − ξ)
[P (r, θ − ξ)] = − P (r, θ − ξ)
∂θ2 1 − 2r cos(θ − ξ) + r2
4r2 sin2 (θ − ξ)
+ P (r, θ − ξ)
(1 − 2r cos(θ − ξ) + r2 )2
4r2 sin2 (θ − ξ)
+ P (r, θ − ξ),
(1 − 2r cos(θ − ξ) + r2 )2
or
∂2 2r cos(θ − ξ)
[P (r, θ − ξ)] = − P (r, θ − ξ)
∂θ2 1 − 2r cos(θ − ξ) + r2
(5.418)
8r2 sin2 (θ − ξ)
+ P (r, θ − ξ).
(1 − 2r cos(θ − ξ) + r2 )2
Taking absolute values on both sides of (5.418) and applying the triangle in-
equality, we obtain
∂2 2r 8r2
2
[P (r, θ − ξ)] 6 2
P (r, θ − ξ) + P (r, θ − ξ), (5.419)
∂θ (1 − r) (1 − r)4
where we have also used the estimate in (5.409) and the positivity of the Poisson
kernel (see property (i) in Proposition 5.3.15). Integrating from −π to π on both
sides of (5.419) then yields
Z π
∂2 2r 8r2
2
[P (r, θ − ξ)] dξ 6 + , for 0 6 r < 1, θ ∈ R,
−π ∂θ (1 − r)2 (1 − r)4
where we have also used property (iv) in Proposition 5.3.15; thus, we have shown
∂2
that [P (r, θ − ξ)] is absolutely integrable over [−π, π] for 0 6 r < 1 and
∂θ2
θ ∈ R.
Next, differentiate the Poisson kernel in (5.408) with respect to r, for 0 6
r < 1, to obtain
∂ 1 −2r 1 (1 − r2 )(2r − 2 cos(θ − ξ))
[P (r, θ − ξ)] = 2
− ,
∂r 2π 1 − 2r cos(θ − ξ) + r 2π (1 − 2r cos(θ − ξ) + r2 )2
where we have applied the Product Rule; so that, in view of the expression for
the Poison kernel in (5.408),
∂ 2r 2r − 2 cos(θ − ξ)
[P (r, θ − ξ)] = − 2
P (r, θ − ξ) − P (r, θ − ξ),
∂r 1−r 1 − 2r cos(θ − ξ) + r2
or

∂ −2r 2r − cos(θ − ξ)
[P (r, θ − ξ)] = + P (r, θ − ξ), (5.420)
∂r 1 − r2 1 − 2r cos(θ − ξ) + r2
for 0 6 r < 1 and all ξ and θ in R.
Next, take absolute values on both sides of (5.420), applying the triangle
inequality, and use the estimate in (5.409) to obtain

∂ 2r 2r + 1
[P (r, θ − ξ)] 6 + P (r, θ − ξ), (5.421)
∂r 1 − r2 (1 − r)2
for 0 6 r < 1 and θ, ξ ∈ R, where we have also used the positivity of the Poisson
kernel (see property (i) in Proposition 5.3.15).
Integrating from −π to π on both sides of (5.421) and using property (iv) in

Proposition 5.3.15 then yields
Z π
∂ 2r 2r + 1
[P (r, θ − ξ)] dξ 6 2
+ , for 0 6 r < 0, θ ∈ R,
−π ∂r 1 − r (1 − r)2
∂
which shows that [P (r, θ − ξ)] is absolutely integrable over [−π, π] for 0 6
∂r
r < 1 and θ ∈ [−π, π].
Hence, differentiation under the integral sign in the first part of the definition
of u in (5.412) is justified. (See the results in Appendix B.1). We have therefore
established that the function v defined in (5.412) is in C 2 (D1 ) and satisfies
Laplace’s equation. It remains to prove that v ∈ C(D1 ) and that it satisfies
the boundary conditions in problem (5.334). This will be accomplished once we
prove the following lemma:
Lemma 5.3.16 (Boundary Limits of the Poisson Integral Representation). Let
v be as given in (5.412) where f is continuous on [−π, π]. Then, for every
ζ ∈ [−π, π],
lim |v(r, θ) − f (ζ)| = 0. (5.422)
(r,θ)→(1,ζ)
Proof: First consider the case in which ζ ∈ (−π, π).

Let ε > 0 be given. Since f is is continuous on [−π, π], there exists δ1 > 0
π
such that δ1 < , and
2
ε
|ξ − ζ| < δ1 ⇒ ξ ∈ (−π, π) and |f (ξ) − f (ζ)| < . (5.423)
3
Next, use property (iii) of the Poisson kernel in Proposition 5.3.15 to write
Z π Z π
v(r, θ) − f (ζ) = P (r, θ − ξ)g(ξ) dξ − f (ζ) P (r, θ − ξ) dξ
−π −π
Z π Z π
= P (r, θ − ξ)f (ξ) dξ − P (r, θ − ξ)f (ζ) dξ,
−π −π
so that
Z π
v(r, θ) − f (ζ) = P (r, θ − ξ)(f (ξ) − f (ζ)) dξ. (5.424)
−π
Next, take absolute values on both sides of (5.424) and use the positivity of the
Poisson kernel (see property (i) in Proposition 5.3.15) to obtain that
Z π
|v(r, θ) − f (ζ)| 6 P (r, θ − ξ)|f (ξ) − f (ζ)| dξ. (5.425)
−π
We’ll next divide the integral on the right–hand side of (5.425) into three inte-
grals over the domains [−π, ζ − δ1 ], [ζ − δ1 , ζ + δ1 ] and [ζ + δ1 , π], respectively.
We first estimate the integral over [ζ − δ1 , ζ + δ1 ] using (5.424) to get

Z ζ+δ1
ε ζ+δ1
Z
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < P (r, θ − ξ) dξ,
ζ−δ1 3 ζ−δ1
so that, by virtue of the positivity of the Poisson (property (i) in Proposition

5.3.15)
Z ζ+δ1
ε π
Z
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < P (r, θ − ξ) dξ;
ζ−δ1 3 −π
hence, by property (iv) in Proposition 5.3.15,

Z ζ+δ1
ε
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < . (5.426)
ζ−δ1 3
Next, we estimate the integral over [ζ + δ1 , π]. Using the estimate in (5.415)
and the triangle inequality we obtain
Z π Z π
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < 2M P (r, θ − ξ) dξ, (5.427)
ζ+δ1 ζ+δ1
for all θ ∈ (−π, π) and 0 6 r < 1. Then, for

δ1
|θ − ζ| < , (5.428)
2
we obtain from (5.427) and the positivity of the Poisson kernel that
Z π Z π
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < 2M P (r, θ − ξ) dξ; (5.429)
ζ+δ1 θ+δ1 /2
Thus, making the change of variables ω = ξ − θ in (5.429) we get

Z π Z π−θ
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < 2M P (r, ω) dω,
ζ+δ1 δ1 /2
so that, by the positivity of the Poisson kernel,

Z π Z π
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < 2M P (r, ω) dω. (5.430)
ζ+δ1 δ1 /2
Now, it follows from the properties of the cosine function that

δ1
< ω < π ⇒ cos(ω) < cos(δ1 /2);
2
so that, for 0 < r < 1,
δ1
< ω < π ⇒ 1 − 2r cos(ω) + r2 > 1 − 2r cos(δ1 /2) + r2 ;
2
so that, by the expression for the Poisson kernel in (5.408),
P (r, ω) < P (r, δ1 /2), for all ω ∈ [δ1 /2, π]. (5.431)

Z π Z π
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < 2M P (r, δ1 /2) dω,
ζ+δ1 δ1 /2
for 0 < r < 1; so that,

Z π
δ1
P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < 2M π − P (r, δ1 /2), (5.432)
ζ+δ1 2
for 0 < r < 1.

Now, it follows property (v) in Proposition 5.3.15 that there exists δ2 > 0
such that
ε
|r − 1| < δ2 ⇒ P (r, δ1 /2) < (5.433)
3M (2π − δ1 )
Thus, combining (5.432), (5.428) and (5.433) we see that
Z π
δ1 ε
|r − 1| < δ2 and |θ − ζ| < ⇒ P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < . (5.434)
2 ζ+δ1 3
Calculations similar to those leading to (5.434) show that there exists δ3 > 0
such that
Z ζ−δ1
δ1 ε
|r − 1| < δ3 and |θ − ζ| < ⇒ P (r, θ − ξ)|f (ξ) − f (ζ)| dξ < . (5.435)
2 −π 3

δ1
Letting δ = min , δ2 , δ3 , we see that, in view of (5.425), (5.426), (5.434)
2
and (5.435),
|r − 1| < δ and |θ − ζ| < δ ⇒ |v(r, θ) − f (ζ)| < ε.
This completes the proof of the boundary limits lemma for the case ζ ∈ (−π, π).
The case in which ζ is one of the end–points of the interval [−π, π] can be treated
in an analogous manner to the interior point case using one–sided limits at those
points.
5.3.6 Existence for the Dirichlet Problem on the Unit Disc

In Section 5.3.5 we studied the Dirichlet problem of the two–dimensional Lapla-
cian in the unit disc in polar coordinates,
 2
∂ v 1 ∂v 1 ∂2v

 + + = 0, 0 < r < 1, −π < θ < π;
∂r2 r ∂r r2 ∂θ2 (5.436)


v(1, θ) = f (θ), −π 6 θ 6 π,
where f : R → R is a continuous, 2π–periodic function. We constructed a

candidate for a solution given by (5.412); namely,
Z π


 P (r, θ − ξ)f (ξ) dξ, for 0 6 r < 1, θ ∈ [−π, π];
−π
v(r, θ) = (5.437)



f (θ), for r = 1, θ ∈ [−π, π],
where P (r, θ) denotes the Poisson kernel for the unit disc in R2 given in (5.396)
or (5.408); that is,
1 1 − r2
P (r, θ) = , for 0 6 r < 1 and θ ∈ [−π, π], (5.438)
2π 1 − 2r cos(θ) + r2
We showed in Section 5.3.5 that
Z π
v(r, θ) = P (r, θ − ξ)f (ξ) dξ, for 0 6 r < 1, θ ∈ [−π, π], (5.439)
−π
is a C 2 function that solves Laplace’s equation in D1 . It remains to show that

v ∈ C(D1 , R) and that v(1, θ) = f (θ), for all θ ∈ [−π, π].
The fact that the function v : D1 → R defined in (5.437) is continuous follows
from Lemma 5.3.16. Indeed, for v(r, θ) as given in (5.439), we get from Lemma
5.3.16 that
lim v(r, θ) = f (ζ), for all ζ ∈ [−π, π];
(r,θ)→(1,ζ)
so that, in view of the definition of v in (5.437),
lim v(r, θ) = v(1, ζ), for all ζ ∈ [−π, π],

(r,θ)→(1,ζ)
which shows that v is continuous at every point (1, ζ) ∈ ∂D1 . This completes
the proof of the following result.
Theorem 5.3.17 (Existence for the Dirichlet Problem for the Disc, I). Let
f : R → R be a continuous, 2π–periodic function. There exists a function v ∈
C 2 (D1 , R) ∩ C(D1 , R) that solves the BVP in (5.436).
By converting back to Cartesian coordinates from polar coordinates, we
obtain a solution of the Dirichlet problem in (5.332); namely,

(5.440)
u(x, y) = g(x, y), for (x, y) ∈ ∂D1 ,
where g ∈ C(∂D1 , R). Indeed, for (x, y) ∈ D1 , write
x = r cos θ and y = r sin θ, for 0 6 r < 1 and − π < θ 6 π,
and, for −π < ξ 6 π, put
ω1 (ξ) = cos ξ and ω2 = sin ξ;

so that, (ω1 (ξ), ω2 (ξ)) ∈ ∂D1 for all ξ ∈ (−π, π]. Consequently, denoting the
Euclidean norm of vector in R2 by | · |,
|(x, y) − (ω1 , ω2 )|2 = r2 − 2(x, y) · (ω1 , ω2 ) + 1, (5.441)
where the middle–term on the right–hand side of (5.441) is the dot–product of
the the vectors (x, y) and (ω1 , ω2 ). Thus,
|(x, y) − (ω1 , ω2 )|2 = r2 − 2r cos(θ − ξ) + 1. (5.442)
Hence, in view of (5.438), we see from (5.442) that
1 1 − |(x, y)|2
P (r, θ − ξ) = , (5.443)
2π |(x, y) − (ω1 , ω2 )|2
for (x, y) ∈ D1 and (ω1 , ω2 ) ∈ ∂D1 .
The right–hand side of the equation in (5.443) gives the Poisson kernel in
Cartesian coordinates. We write
1 1 − |(x, y)|2
P ((x, y), (ω1 , ω2 )) = , (5.444)
2π |(x, y) − (ω1 , ω2 )|2
for (x, y) ∈ D1 and (ω1 , ω2 ) ∈ ∂D1 . Then, in view of (5.439) and (5.444), we
define
I
u(x, y) = P ((x, y), ω)g(ω) dsω , for (x, y) ∈ D1 , (5.445)
∂D1
where ω = (ω1 , ω2 ) ∈ ∂D1 and dsω is the element of arc–length along the circle
∂D1 .
The boundary limit Lemma 5.3.16 implies that
I
lim P ((x, y), ω)g(ω) dsω = g(z1 , z2 ), (5.446)
(x,y)→(z1 ,z2 ) ∂D1
(x,y)∈D1
for all (z1 , z2 ) ∈ ∂D1 . Thus, in view of (5.437), (5.445) and (5.446), we see that
the function u : D1 → R given by
I


 P ((x, y), ω)g(ω) dsω , for (x, y) ∈ D1 ;
∂D1
u(x, y) = (5.447)


for (x, y) ∈ ∂D1 ,

g(x, y),
where P ((x, y), ω), for (x, y) ∈ D1 and ω ∈ ∂D1 , is the Poisson kernel for the
unit disc, D1 , given in (5.444), gives a solution of the boundary value problem
in (5.440). We state this fact as the following existence theorem, which is,
essentially, the Cartesian coordinates version of Theorem 5.3.17.
Theorem 5.3.18 (Existence for the Dirichlet Problem for the Disc, II). For
any given g ∈ C(∂D1 , R), there exists a function u ∈ C 2 (D1 , R) ∩ C(D1 , R)
that solves the BVP in (5.440). Indeed, u is given by the Poisson integral
representation in (5.447).
5.4 Green’s Functions

In Section 5.3.6 we constructed a solution of the Dirichlet problem for Laplace’s
equation in the unit disc, D1 , in R2 :

u(x, y) = g(x, y), for (x, y) ∈ ∂D1 ,
where g ∈ C(∂D1 , R) is given. The construction was based on the Poisson

integral representation of a harmonic functions in terms of its boundary values.
In this section, we turn to the Dirichlet problem for Laplace’s equation in a
general bounded domain Ω ⊂ R2 with piecewise smooth boundary ∂Ω:

∆u = 0 in Ω;
(5.448)
u(x, y) = g(x, y), for (x, y) ∈ ∂Ω,
where g ∈ C(∂Ω, R) is given. We will be able to construct a solution of the

Dirichlet problem in (5.448) by using the integral representation of a harmonic
function in Ω in terms of its values on the boundary of Ω. This integral repre-
sentation in turn relies on the existence of a special function tied to the domain
Ω that is known as the Green’s function of the domain.
We will also construct a solution of the Dirichlet problem for Poisson’s equa-
tions:
−∆u(x, y) = f (x, y), for (x, y) ∈ Ω;
(5.449)
u(x, y) = g(x, y), for (x, y) ∈ ∂Ω,
where g ∈ C(∂Ω, R) and f : Ω → R is assumed to be Hölder continuous.
Definition 5.4.1 (Hölder Continuity). Let Ω denote an open subset of R2 and
f : Ω → R be a real valued function. The function f is said to be Hölder
continuous with Hölder exponent α, where 0 < α 6 1, if and only if there
exists a positive constant M such that
|f (x, y) − f (ξ, η)| 6 M |(x, y) − (ξ, η)|α , for all (x, y), (ξ, η) ∈ Ω. (5.450)
If α = 1 in (5.450), we say that f is Lipschitz continuous in Ω, with Lipschitz

constant M .
5.4.1 Green’s Integral Representation Formula

In Section 4.2.1 we saw that the functions of the form
p
W (x, y) = C ln x2 + y 2 , for (x, y) 6= (0, 0), (5.451)
where C is a constant, are harmonic in the punctured plane, R2 \{(0, 0)}; that
is, W solves
uxx + uyy = 0 in R2 \{(0, 0)}. (5.452)
We also saw that the functions in (5.451) are radially symmetric solutions of
the problem in (5.452).
5.4. GREEN’S FUNCTIONS 171
Since Laplace’s equation in (5.452) is also translation invariant, given any

point (xo , yo ) ∈ R2 , it follows that the function
(x, y) 7→ W (x − xo , y − yo ), for (x, y) 6= (xo , yo ), (5.453)
solves Laplace’s equation in R2 \{(xo , yo )}. Denoting the function defined in
(5.453) by W(xo ,yo ) , we see that
p
W(xo ,yo ) (x, y) = C ln (x − xo )2 + (y − yo )2 , for (x, y) 6= (xo , yo ), (5.454)
solves
uxx + uyy = 0 in R2 \{(xo , yo )}. (5.455)
2
Let U denote an open subset of R and let Ω be a connected, bounded and
open subset of U such that Ω ⊂ U. Assume that the boundary, ∂Ω, of Ω is
piecewise C 1 . Let u ∈ C 2 (Ω, R) ∩ C(Ω, R) and v ∈ C 2 (Ω, R) ∩ C(Ω, R). Using
the Divergence Theorem (or Green’s Theorem), we can derive the following
identity ZZ I
∂u ∂v
(v∆u − u∆v) dxdy = v −u ds, (5.456)
Ω ∂Ω ∂n ∂n
where ∆u and ∆v denote the Laplacian of u and v, respectively, and
∂u ∂v
= ∇u · n
b and = ∇v · n
b (5.457)
∂n ∂n
denote the directional derivative of u and v, respectively, in the direction of the
outward, unit normal vector, n b, to the boundary, ∂Ω, of Ω. The differential ds in
the integral on the right–hand side of (5.456) denotes the element of arc–length
along the curve ∂Ω. The identity in (5.456) is called Green’s Identity II.
Next, let (xo , yo ) ∈ Ω and let ε > 0 be such that the closure of the disc of
radius ε centered at (xo , yo ) is contained entirely in Ω. Setting
Dε (xo , yo ) = {(x, y) ∈ R2 | (x − xo )2 + (y − yo )2 < ε2 },
we have that
Dε (xo , yo ) ⊂ Ω.
Define Ωε = Ω\Dε (xo , yo ), the set Ω with Dε (xo , yo ) taken out. Observe that
the function W(xo ,yo ) defined in (5.454) is harmonic in Ωε by virtue of (5.455).
Next, let u ∈ C 2 (U, R) and apply Green’s Identity II in (5.456) with Ωε in
place of Ω, and with W(xo ,yo ) in place of v to get

∂W(xo ,yo )
ZZ I
∂u
W(xo ,yo ) ∆u dxdy = W(xo ,yo ) −u ds, (5.458)
Ωε ∂Ωε ∂n ∂n
where we have used the fact that W(xo ,yo ) is harmonic in Ωε .
The boundary of Ωε is the union of ∂Ω and ∂Dε (xo , yo ). We can therefore
rewrite the boundary integral on the right–hand side of (5.458) as
I I
∂u ∂W ∂u ∂W
W −u ds − W −u ds, (5.459)
∂Ω ∂n ∂n ∂Dε ∂n ∂n
where we have written W for W(xo ,yo ) and Dε for Dε (xo , yo ), and the minus
sign before the second integral in (5.459) indicates the fact that the outward,
unit normal of ∂Ωε at a point on ∂Dε (xo , yo ), points toward the point (xo , yo ).
We can therefore rewrite (5.458) as
ZZ I I
∂u ∂W ∂u ∂W
W ∆u dxdy = W −u ds − W −u ds,
Ωε ∂Ω ∂n ∂n ∂Dε ∂n ∂n
which can be rewritten as

I ZZ
∂u ∂W
W −u ds = − W ∆u dxdy
∂Dε ∂n ∂n Ωε
I (5.460)
∂u ∂W
+ W −u ds.
∂Ω ∂n ∂n
We proceed to compute the limit as ε → 0+ of the integrals in (5.460) that

depend on ε. These calculations will depend on the definition of W in (5.454)
and the assumption that u ∈ C 2 (U, R). We begin by showing that
ZZ ZZ
lim W ∆u dxdy = W ∆u dxdy. (5.461)
ε→0+ Ωε Ω
To establish (5.461), first note that

ZZ ZZ ZZ
W ∆u dxdy − W ∆u dxdy = W ∆u dxdy;
Ω Ωε Dε
so that,
ZZ ZZ ZZ
W ∆u dxdy − W ∆u dxdy 6 |W ||∆u| dxdy. (5.462)
Ω Ωε Dε
Now, since we are assuming that u ∈ C 2 (U, R) and Ω is bounded, |∆u| is

bounded on Ω. Putting
M = max |∆u(x, y)|,

(x,y)∈Ω
we obtain from (5.462) that

ZZ ZZ ZZ
W ∆u dxdy − W ∆u dxdy 6 M |W (x, y)| dxdy, (5.463)
Ω Ωε Dε
where W is defined in (5.454). We can use polar coordinates to evaluate the

integral on the right–hand side of (5.463) as follows
ZZ Z 2π Z ε
|W (x, y)| dxdy = |C| | ln r|r drdθ.
Dε 0 0
Thus, assuming that ε < 1,

ZZ Z ε
|W (x, y)| dxdy = −2π|C| r ln r dr. (5.464)
Dε 0
We can use integration by parts and L’Hospital’s Rule to evaluate the integral
on the right–hand side of (5.464) to get
2
ε2
ZZ
ε
|W (x, y)| dxdy = −2π|C| ln ε − . (5.465)
Dε 2 4
Thus, combining the estimate in (5.463) with the result in (5.465),

ZZ ZZ 2
ε
W ∆u dxdy − W ∆u dxdy 6 πM |C| + ε|ε ln ε| , (5.466)
Ω Ωε 2
for 0 < ε < 1. Letting ε → 0+ in (5.466) and using the fact that
lim ε ln ε = 0, (5.467)
ε→0+
which follows from L’Hospital’s Rule, we obtain from (5.466) and (5.467) that
ZZ ZZ
lim+ W ∆u dxdy − W ∆u dxdy = 0,
ε→0 Ωε Ω
which implies (5.461).

Next, we estimate the first boundary integral on the left–hand side of (5.460)
to get I I
∂u
W ds 6 M1 |W | ds, (5.468)
∂Dε ∂n ∂Dε
where M1 is given by
M1 = max |∇u(x, y)|,
(x,y)∈Ω
according to the definition of the outward–normal derivative of u in (5.457).

Using the definition of W in (5.454) we can evaluate the integral on the
right–hand side of (5.468) to obtain
I
|W | ds = 2πε| ln ε|,
∂Dε
for 0 < ε < 1; so that,

I
∂u
W ds 6 2πM1 |ε ln ε|, for 0 < ε < 1. (5.469)
∂Dε ∂n
Consequently, using (5.467) again, we obtain from (5.469) that

I
∂u
lim+ W ds = 0. (5.470)
ε→0 ∂Dε ∂n
It remains to evaluate
I
∂W
lim u ds,
ε→0+ ∂Dε ∂n
∂W
where the outward, normal derivative, , of W given in (5.454) is the deriva-
∂n
tive of W with respect to the outward, radial direction from (xo , yo ) on the
boundary of Dε (xo , yo ). Wring
p
r = (x − xo )2 + (y − yo )2 ,
we then have that

∂W d C
= [C ln r] = ;
∂n ∂Dε (xo ,yo ) dr r=ε ε
so that, I I
∂W C
u ds = u ds;
∂Dε ∂n ε ∂Dε
or I I
∂W 1
u ds = 2πC u ds, for 0 < ε < 1, (5.471)
∂Dε ∂n 2πε ∂Dε
I
1
where u ds is the mean value (or average value) of u on ∂Dε (xo , yo ).
2πε ∂Dε
Now, since u is continuous in U, which contains Ω, it follows that
I
1
lim u ds = u(xo , yo ). (5.472)
ε→0+ 2πε ∂Dε
Thus, combining (5.471) and (5.472), we get that

I
∂W
lim u ds = 2πCu(xo , yo ). (5.473)
ε→0 +
∂Dε ∂n
Next, let ε → 0+ on both sides of (5.460), and use the limits in (5.461),
(5.470) and (5.473), to get that
ZZ I
∂u ∂W
−2πCu(xo , yo ) = − W ∆u dxdy + W −u ds. (5.474)
From this point on we set
−2πC = 1,
or
1
C=− ; (5.475)
2π
so that, the equation in (5.474) becomes
ZZ I
∂u ∂W
u(xo , yo ) = − W ∆u dxdy + W −u ds, (5.476)
where, according to (5.454) and (5.454), W represents the function
W(xo ,yo ) : R2 \{(xo , yo )} → R
defined by
1 p
W(xo ,yo ) (x, y) = − ln (x − xo )2 + (y − yo )2 , if (x, y) 6= (xo , yo ). (5.477)
2π
It what follows we will show how to use the formula in (5.476) to obtain a
representation of harmonic function u ∈ C 2 (Ω, R)∩C(Ω, R) in terms of its values
on the boundary of Ω. If we succeed, we will be able to obtain a representation
for a solution of the Dirichlet problem in (5.448); namely,

uxx + uyy = 0 in Ω;
(5.478)
u(x, y) = g(x, y), for (x, y) ∈ ∂Ω,
where g : ∂Ω → R is a given continuous function on ∂Ω.

To obtain an integral representation for u ∈ C 2 (U, R) in terms of its values
on ∂Ω, let H ∈ C 2 (U, R) be a function that is harmonic in Ω; so that,
∆H = 0, in Ω. (5.479)
Next, apply Green’s Identity II in (5.456), with u ∈ C 2 (U, R) and H in place of

v, to get ZZ I
∂u ∂H
H∆u dxdy = H −u ds,
where we have used (5.479), which we can rewrite as
ZZ I
∂u ∂H
0=− H∆u dxdy + H −u ds. (5.480)
Adding the expressions in (5.476) and (5.480) we obtain
ZZ
u(xo , yo ) = − (W + H)∆u dxdy
Ω
I I (5.481)
∂u ∂
+ (W + H) ds − u [W + H] ds.
∂Ω ∂n ∂Ω ∂n
Suppose that we can find a C 2 function H : Ω → R that is harmonic in Ω

and such that
W + H = 0 on ∂Ω,
or
H = −W on ∂Ω. (5.482)
ZZ I
∂
u(xo , yo ) = − (W + H)∆u dxdy − u [W + H] ds, (5.483)
Ω ∂Ω ∂n
for any (xo , yo ) ∈ Ω, where W is given by (5.477), and H ∈ C 2 (Ω, R) ∩ C(Ω, R)

is a harmonic function in Ω satisfying the boundary condition in (5.482).
Since (xo , yo ) ∈ Ω in (5.483) is arbitrary, by defining
1 p
W(x,y) (ξ, η) = − ln (ξ − x)2 + (η − y)2 , if (ξ, η) 6= (x, y), (5.484)
2π
and (x, y) ∈ Ω, we obtain from (5.483) the integral representation formula
ZZ
u(x, y) = − (W(x,y) (ξ, η) + H(ξ, η))∆u(ξ, η) dξdη
Ω
I (5.485)
∂
− u(ξ, η) [W(x,y) (ξ, η) + H(ξ, η)] ds,
∂Ω ∂n
for all (x, y) ∈ Ω.

We shall refer to the expression in (5.485) as Green’s Integral Represen-
tation Formula for a function u ∈ C 2 (Ω, R) ∩ C(Ω, R).
5.4.2 Definition of Green’s Function

We will denote the expression
W(x,y) (ξ, η) + H(ξ, η), for (ξ, η) 6= (x, y), (5.486)
in Green’s Integral Representation Formula in (5.485), where W(x,y) (ξ, η) is as

given in (5.484), by
G((x, y), (ξ, η)) (5.487)
and call it the Green’s function of the domain Ω. With this notation, the Green’s
Integral Representation Formula in (5.485) becomes
ZZ
u(x, y) = − G((x, y), (ξ, η))∆u(ξ, η) dξdη
Ω
I (5.488)
∂
− u(ξ, η) [G((x, y), (ξ, η))] ds,
∂Ω ∂n
for all (x, y) ∈ Ω. Thus, according to (5.488), if we are able to find a Green’s
function for a domain Ω ⊂ R2 , we will be able to represent any function u ∈
C 2 (Ω, R) ∩ C(Ω, R) in terms of its Laplacian, ∆u, in Ω, and its boundary values
on ∂Ω. In other words, we will be able to construct a candidate for a solution
of the Dirichlet problem for Poisson’s equation in (5.449); namely,
ZZ
u(x, y) = G((x, y), (ξ, η))f (ξ, η) dξdη
Ω
I (5.489)
∂
− g(ξ, η) [G((x, y), (ξ, η))] ds,
∂Ω ∂n
for all (x, y) ∈ Ω, provided that we make additional assumptions on f . In

particular, we will assume that f is Hölder continuous with Hölder exponent α,
with 0 < α 6 1 (see Definition 5.4.1 on page 170 in these notes).
For the special case in which u ∈ C 2 (Ω, R) ∩ C(Ω, R) is harmonic in Ω and
u takes on values given by g on ∂Ω, we obtain from (5.488) that
I
∂
u(x, y) = − g(ξ, η) [G((x, y), (ξ, η))] ds, for (x, y) ∈ Ω,
∂Ω ∂n

I
∂
u(x, y) = g(ξ, η) − [G((x, y), (ξ, η))] ds, for (x, y) ∈ Ω. (5.490)
∂Ω ∂n
Thus, setting
∂
P ((x, y), (ξ, η)) = − [G((x, y), (ξ, η)) (5.491)
∂n
in (5.490), we recover the Poisson integral representation formula for a harmonic
function in Ω in terms of of its boundary values,
I
u(x, y) = g(ξ, η)P ((x, y), (ξ, η)) ds, for (x, y) ∈ Ω, (5.492)
∂Ω
that we obtained in Section 5.3.6 for the unit disc, D1 , in R2 .

In this section, we discuss some properties of the Green’s function. We
will then compute the Green’s function for the disc DR in R2 , for R > 0.
In subsequent sections, we will compute the Green’s functions of other simple
regions in the plane.
According to the definition of Green’s function given in (5.487) and (5.486)
at the start of this section, if a domain Ω ⊂ R2 has a Green’s function,
G((x, y), (ξ, η)), for (x, y), (ξ, η) ∈ Ω, with (x, y) 6= (ξ, η), then the function
(ξ, η) 7→ G((x, y), (ξ, η)) − W(x,y) (ξ, η), for (ξ, η) ∈ Ω,
is harmonic in Ω and
G((x, y), (ξ, η)) = 0, for (ξ, η) ∈ ∂Ω.
We note that a given domain Ω can have at most one Green function. For
suppose that there are two functions, G1 ((x, y), (ξ, η)) and G2 ((x, y), (ξ, η)),
with the properties that the maps
(ξ, η) 7→ Gi ((x, y), (ξ, η)) − W(x,y) (ξ, η), for (ξ, η) ∈ Ω, and i = 1, 2,
are harmonic in Ω, and
Gi ((x, y), (ξ, η)) = 0, for (ξ, η) ∈ ∂Ω, and i = 1, 2.

Then, the functions
Hi (ξ, η) = Gi ((x, y), (ξ, η)) − W(x,y) (ξ, η), for (ξ, η) ∈ Ω, and i = 1, 2,
solve the Dirichlet problem

uxx + uyy = 0 in Ω;
(5.493)
u(x, y) = −W(x,y) , on ∂Ω,
where −W(x,y) is continuous on ∂Ω, since (x, y) 6∈ ∂Ω (see the definition of

W(x,y) in (5.484)). Consequently, since the Dirichlet problem in (5.493) can
have at most one solutions (see Problem 5 in Assignment #5), it follows that
H1 (ξ, η) = H2 (ξ, η), for all (ξ, η) ∈ Ω.
Consequently,
G1 ((x, y), (ξ, η)) = G2 ((x, y), (ξ, η)), for all (ξ, η) ∈ Ω with (ξ, η) 6= (x, y).
In the following example, we compute the Green’s function for the disc of
radius R around the origin, DR , in R2 .
Example 5.4.2 (Green’s Function of DR ). For any (x, y) ∈ DR , we construct

a function H(x,y) : DR → R that is harmonic in Ω and such that
H(x,y) (ξ, η) = −W(x,y) (ξ, η), for all (ξ, η) ∈ ∂DR . (5.494)
In the construction, we will use the fact that, if (x∗ , y ∗ ) 6∈ DR , then the function
p
(ξ, η) 7→ ln (ξ − x∗ )2 + (η − y ∗ )2 , for (ξ, η) ∈ DR ,
or
(ξ, η) 7→ ln |(ξ, η) − (x∗ , y ∗ )|, for (ξ, η) ∈ DR ,
is harmonic in DR .
We first consider the case in which (x, y) ∈ DR and (x, y) 6= (0, 0); the case
(x, y) = (0, 0) with be dealt with separately.
Assume that (x, y) ∈ DR and (x, y) 6= (0, 0). We pick (x∗ , y ∗ ) to be a point
on the ray emanating from the origin, (0, 0), and going through the point (x, y)
at a distance λ > R from the origin; so that,
λ
(x∗ , y ∗ ) = (x, y), (5.495)
|(x, y)|
where λ will be chosen shortly.

We consider the function H(x,y) : DR → R given by
1
H(x,y) (ξ, η) = ln (Q(x, y)|(ξ, η) − (x∗ , y ∗ )|) , for (ξ, η) ∈ DR , (5.496)
2π
Where Q(x, y) > 0, along with λ > R in (5.495), be chosen so that the boundary
condition in (5.494) is satisfied.
We note that H(x,y) is harmonic in DR , since |(x∗ , y ∗ )| = λ > R, according
to (5.495).
Next, use the definition of (x∗ , y ∗ ) in (5.495) to compute, for (ξ, η) ∈ ∂DR ,
λ
|(ξ, η) − (x∗ , y ∗ )|2 = R2 − 2 (x, y) · (ξ, η) + λ2 ,
|(x, y)|
or
R2 |(x, y)|

λ
|(ξ, η) − (x∗ , y ∗ )|2 = − 2(x, y) · (ξ, η) + λ|(x, y)| . (5.497)
|(x, y)| λ
Setting λ|(x, y)| = R2 in (5.497); so that,
R2
λ= , (5.498)
|(x, y)|
we obtain from (5.497) that
R2
|(ξ, η) − (x∗ , y ∗ )|2 = |(x, y)|2 − 2(x, y) · (ξ, η) + R2 ,

|(x, y)|2
or
R2
|(ξ, η) − (x∗ , y ∗ )|2 = |(x, y) − (ξ, η)|2 ,
|(x, y)|2
|(x, y)|
|(ξ, η) − (x∗ , y ∗ )| = |(ξ, η) − (x, y)|, for (ξ, η) ∈ ∂DR . (5.499)
R

1 |(x, y)| 1
ln |(ξ, η) − (x∗ , y ∗ )| = ln (|(ξ, η) − (x, y)|) , (5.500)
2π R 2π
for all (ξ, η) ∈ ∂DR . Thus, comparing (5.496) with the left–hand side of (5.500)
we see that, we see that we can take
|(x, y)|
Q(x, y) =
R
in (5.496) to obtain

1 |(x, y)|
H(x,y) (ξ, η) = ln |(ξ, η) − (x∗ , y ∗ )| , for (ξ, η) ∈ DR , (5.501)
2π R
where, according to (5.495) and (5.498),
R2
(x∗ , y ∗ ) = (x, y), for (x, y) ∈ DR with (x, y) 6= (0, 0). (5.502)
|(x, y)|2
The identity in (5.500) then shows that
H(x,y) (ξ, η) = −W(x,y) (ξ, η), for (ξ, η) ∈ DR ,
which is the assertion in (5.494). It then follows from the definition of the Green
function in (5.486) and (5.487) that
G((x, y), (ξ, η)) = H(x,y) (ξ, η) + W(x,y) (ξ, η),
for (x, y), (ξ, η) ∈ DR with (x, y) 6= (ξ, η), and (x, y) 6= (0, 0), or

1 |(x, y)| ∗ ∗
G((x, y), (ξ, η)) = ln |(ξ, η) − (x , y )|
2π R
(5.503)
1
− ln (|(ξ, η) − (x, y)|) ,
2π
for (x, y), (ξ, η) ∈ DR with (x, y) 6= (ξ, η), and (x, y) 6= (0, 0).
Next, for the case (x, y) = (0, 0), observe that the constant function
1
H(ξ, η) = ln R, for (ξ, η) ∈ R2 ,
2π
is harmonic everywhere. Furthermore, setting
1
W (ξ, η) = − ln |(ξ, η)|, for (ξ, η) 6= (0, 0),
2π
we see that
H(ξ, η) = −W (ξ, η), for all (ξ, η) ∈ ∂DR .
Thus, we define
1 1
G((0, 0), (ξ, η)) = ln R − ln |(ξ, η)|, for (ξ, η) 6= (0, 0). (5.504)
2π 2π
The function, G((x, y), (ξ, η)), defined in (5.503) and (5.504) is the Green
function of the disc of radius R around the origin in R2 . We can use it to solve
the following Dirichlet problem for Poisson’s equation in DR :

−∆u(x, y) = f (x, y), for (x, y) ∈ DR ;
(5.505)
u(x, y) = g(x, y), for (x, y) ∈ ∂DR ,
where g ∈ C(∂DR , R) and f : DR → R is assumed to be Hölder continuous.

Indeed, according to (5.489), a solution of (5.505) should satisfy
ZZ
DR
I (5.506)
∂
+ g(ξ, η) − [G((x, y), (ξ, η))] ds,
∂DR ∂n
for all (x, y) ∈ DR .

Next, we compute the normal derivative of G((x, y), (ξ, η)) in the last integral
in (5.506), which is
∂ 1
[G((x, y), (ξ, η))] = ∇(ξ,η) G((x, y), (ξ, η)) · (ξ, η), (5.507)
∂n R
for (ξ, η) ∈ ∂DR , where ∇(ξ,η) denotes the gradient take with respect to the
variables ξ and η.
For (x, y) ∈ DR and (x, y) 6= (0, 0), use (5.503) to write
1 |(x, y)| 1
G((x, y), (ξ, η)) = ln + ln |(ξ, η) − (x∗ , y ∗ )|2
2π R 4π
(5.508)
1
− ln |(ξ, η) − (x, y)|2 ;
4π
so that, taking the partial derivative with respect to ξ,
∂ 1 ξ − x∗
G((x, y), (ξ, η)) =
∂ξ 2π |(ξ, η) − (x∗ , y ∗ )|2
(5.509)
1 ξ−x
− ,
2π |(ξ, η) − (x, y)|2
for (ξ, η) ∈ ∂DR .

Similarly, taking the partial with respect to η on both sides of (5.508),
∂ 1 η − y∗
G((x, y), (ξ, η)) =
∂η 2π |(ξ, η) − (x∗ , y ∗ )|2
(5.510)
1 η−y
− ,
2π |(ξ, η) − (x, y)|2
for (ξ, η) ∈ ∂DR .

We can now use (5.509) and (5.510) to compute (5.507) to get
∂ 1 ξ 2 + η 2 − x∗ ξ − y ∗ η
G((x, y), (ξ, η)) =
∂n 2πR |(ξ, η) − (x∗ , y ∗ )|2
1 ξ 2 + η 2 − xξ − yη
− ,
2πR |(ξ, η) − (x, y)|2
for (ξ, η) ∈ ∂DR , or
∂ 1 R2 − (x∗ , y ∗ ) · (ξ, η)
G((x, y), (ξ, η)) =
∂n 2πR |(ξ, η) − (x∗ , y ∗ )|2
(5.511)
1 R2 − (x, y) · (ξ, η)
− ,
2πR |(ξ, η) − (x, y)|2
for (ξ, η) ∈ ∂DR .

Next, use the definition of (x∗ , y ∗ ) in (5.502) and the identity in (5.499) to
rewrite (5.511) as
∂ 1 |(x, y)|2 − (x, y) · (ξ, η)
G((x, y), (ξ, η)) =
∂n 2πR |(ξ, η) − (x, y)|2
1 R2 − (x, y) · (ξ, η)
− ,
2πR |(ξ, η) − (x, y)|2
for (ξ, η) ∈ ∂DR , which can be rewritten as
∂ 1 |(x, y)|2 − R2
G((x, y), (ξ, η)) = , for (ξ, η) ∈ ∂DR ,
∂n 2πR |(ξ, η) − (x, y)|2
∂ 1 R2 − |(x, y)|2
− G((x, y), (ξ, η)) = , for (ξ, η) ∈ ∂DR . (5.512)
∂n 2πR |(ξ, η) − (x, y)|2
Observe that the expression in (5.512) also works for (x, y) = (0, 0), according
to the definition of G((0, 0), (ξ, η)) in (5.504).
The expression on the right–hand side of (5.512) is the Poisson kernel,
P ((x, y), (ξ, η)), for the disc DR (compare with the result of Problem 2 in As-
signment #6). We therefore get from (5.506) and (5.512) that the solution of
the Dirichlet problem for Poisson’s equation in (5.505) should have the repre-
sentation
ZZ
DR
I (5.513)
+ P ((x, y), (ξ, η))g(ξ, η) ds,
∂DR
for all (x, y) ∈ DR , where

1 R2 − |(x, y)|2
P ((x, y), (ξ, η)) = , for (ξ, η) ∈ ∂DR , (5.514)
2πR |(ξ, η) − (x, y)|2
and (x, y) ∈ DR , is the Poisson kernel of the disc DR , and G((x, y), (ξ, η)) is its
Green’s function defined in (5.503) and (5.504).
Next, we consider two special cases of the representation formula in (5.513).
First, suppose that f (x, y) = 0 for all (x, y) ∈ DR in (5.513) and (5.505);
then, I
u(x, y) = P ((x, y), (ξ, η))g(ξ, η) ds, (5.515)
∂DR
for all (x, y) ∈ DR , where P ((x, y), (ξ, η)) is the Poisson kernel of DR given in
(5.514) solves the Dirichlet problem

∆u = 0, in DR ;
u(x, y) = g(x, y), for (x, y) ∈ ∂DR ,
where g ∈ C(∂DR , R) is a given continuous function. We have therefore recov-

ered the Poisson integral representation of a harmonic function in (5.515) that
we derived in Section 5.3 for the unit disc D1 using separation of variables and
Fourier series.
Second, suppose that g(x, y) = 0 for all (x, y) ∈ ∂DR in (5.513) and (5.505);
then,
ZZ
u(x, y) = G((x, y), (ξ, η))f (ξ, η) dξdη, for (x, y) ∈ DR , (5.516)
DR
where G((x, y), (ξ, η)) is the Green’s function of DR defined in (5.503) and
(5.504), is a candidate for a solution of the Dirichlet problem for Poisson’s
equation in DR :

−∆u(x, y) = f (x, y), for (x, y) ∈ DR ;
(5.517)
u(x, y) = 0, for (x, y) ∈ ∂DR ,
provided that we assume that f : DR → R is Hölder continuous in Ω. We will

show this in the next section.
5.4.3 Solving Poisson’s Equation

The goal of this section is to show that, if f : DR → R is Hölder continuous with
Hölder exponent α ∈ (0, 1] (see Definition 5.4.1), then then function u given in
(5.516) solves the Dirichlet problem for Poisson’s equation in (5.517).
We will frame the discussion in this section in the context of a general do-
main Ω, with piecewise C 1 boundary, ∂Ω, and for which a Green’s function,
G((x, y), (ξ, η)), exists. An instance of such a domain is the disc of radius R
around the origin in R2 , DR , discussed in the previous section. Thus, we will
show that the function u : Ω → R given by
ZZ
u(x, y) = G((x, y), (ξ, η))f (ξ, η) dξdη, for (x, y) ∈ Ω, (5.518)
Ω
where f : Ω → R is a Hölder continuous function with Hölder exponent α ∈

(0, 1], solves the Dirichlet problem

−∆u(x, y) = f (x, y), for (x, y) ∈ Ω;
(5.519)
u(x, y) = 0, for (x, y) ∈ ∂Ω.
Before we prove this result, we will have to establish an additional property

of the Green function.
Let Ω denote an open, bounded subset of R2 with piecewise C 1 boundary,
∂Ω, and suppose that Ω has a Green function, G((x, y), (ξ, η)), for (x, y), (ξ, η) ∈
Ω such that (x, y) 6= (ξ, η). We have already seen that a Green function for Ω,
if it exists, it is unique. We have also seen that
G((x, y), (ξ, η)) = H((x, y), (ξ, η)) + W ((x, y), (ξ, η)), (5.520)
where the map
(ξ, η) 7→ H((x, y), (ξ, η)), for (ξ, η) ∈ Ω,
is harmonic if Ω; so that,
Hξξ + Hηη = 0, in Ω, (5.521)
and
1
W ((x, y), (ξ, η)) = − ln |(ξ, η) − (x, y)|, for (x, y) 6= (ξ, η). (5.522)
2π
It is also the case that
G((x, y), (ξ, η)) = 0, for all (x, y) ∈ Ω and (ξ, η) ∈ ∂Ω. (5.523)
We will next show that G is symmetric; that is,
G((x1 , y1 ), (x2 , y2 )) = G((x2 , y2 ), (x1 , y1 )), for (x1 , y1 ) 6= (x2 , y2 ), (5.524)
and (x1 , y1 ), (x2 , y2 ) ∈ Ω.

To establish (5.524), we proceed as in the derivation of the Green Represen-
tation Formula in Section 5.4.1.
Assume that (x1 , y1 ), (x2 , y2 ) ∈ Ω are such that
(x1 , y1 ) 6= (x2 , y2 ). (5.525)
Define a function v1 : Ω\{(x1 , y1 )} → R by
v1 (ξ, η) = G((x1 , y1 ), (ξ, η)), for (ξ, η) ∈ Ω\{(x1 , y1 )}. (5.526)
It then follows from the definition of the Green’s function in (5.520), (5.521)
and (5.522) that v1 is harmonic in Ω\{(x1 , y1 )}; indeed, there exists a harmonic
function h1 : Ω → R such that
v1 (ξ, η) = h1 (ξ, η) + W ((x1 , y1 ), (ξ, η)) for (ξ, η) ∈ Ω\{(x1 , y1 )}. (5.527)
Furthermore,
v1 (ξ, η) = 0, for (ξ, η) ∈ ∂Ω, (5.528)
by virtue of (5.523).
Similarly, defining v2 : Ω\{(x2 , y2 )} → R by
v2 (ξ, η) = G((x2 , y2 ), (ξ, η)), for (ξ, η) ∈ Ω\{(x2 , y2 )}, (5.529)
there exists a harmonic function h2 : Ω → R such that
v2 (ξ, η) = h2 (ξ, η) + W ((x2 , y2 ), (ξ, η)), for (ξ, η) ∈ Ω\{(x2 , y2 )}; (5.530)
so that, v2 is harmonic in Ω\{(x2 , y2 )} and
v2 (ξ, η) = 0, for (ξ, η) ∈ ∂Ω. (5.531)

We will show that

v1 (x2 , y2 ) = v2 (x1 , y1 ). (5.532)
This will establish (5.524), in view of the definitions of v1 and v2 in (5.526) and
(5.529), respectively.
Since Ω is open and (x1 , y1 ) and (x2 , y2 ) are distinct points in Ω (see (5.525),
there exist εo > 0 such that
Dεo (x1 , y1 ) ⊂ Ω, Dεo (x2 , y2 ) ⊂ Ω,
and
Dεo (x1 , y1 ) ∩ Dεo (x2 , y2 ) = ∅.
We may assume that εo < 1.
Let 0 < ε < εo . As in the derivation of the Green Representation Formula
in Section 5.4.1, define
Ωε = Ω\(Dε (x1 , y1 ) ∪ Dε (x2 , y2 )),
and observe that, by virtue of the definitions of v1 and v2 in (5.527) and (5.530),
respectively, v1 and v2 are harmonic in Ωε . Thus, applying Green’s Identity II
in (5.456) with v1 and v2 in place of u and v, and Ωε in place of Ω,
I ZZ
∂v1 ∂v2
v2 − v1 ds = (v2 ∆v1 − v1 ∆v2 ) dξdη,
∂Ωε ∂n ∂n Ωε

I
∂v1 ∂v2
v2 − v1 ds = 0, (5.533)
∂Ωε ∂n ∂n
since v1 and v2 are harmonic in Ωε .
Observe that the boundary of Ωε is made up of the boundary of Ω together
with the circles ∂Dε (x1 , y1 ) and ∂Dε (x2 , y2 ). We can therefore the integral on
the left–hand side as
I I
∂v1 ∂v2 ∂v1 ∂v2
v2 − v1 ds = v2 − v1 ds
∂Ωε ∂n ∂n ∂Ω ∂n ∂n
I
∂v1 ∂v2
− v2 − v1 ds
∂D1 ∂n ∂n
I
∂v1 ∂v2
− v2 − v1 ds,
∂D2 ∂n ∂n
or, in view of (5.528) and (5.531),
I I
∂v1 ∂v2 ∂v1 ∂v2
v2 − v1 ds = − v2 − v1 ds
∂Ωε ∂n ∂n ∂D1 ∂n ∂n
I (5.534)
∂v1 ∂v2
− v2 − v1 ds,
∂D2 ∂n ∂n
where we have written D1 for Dε (x1 , y1 ) and D2 for Dε (x2 , y2 ). Combining

(5.533) and (5.534) we then get that
I I
∂v1 ∂v2 ∂v1 ∂v2
v2 − v1 ds = − v2 − v1 ds,
∂D2 ∂n ∂n ∂D1 ∂n ∂n
or
I I
∂v1 ∂v2 ∂v2 ∂v1
v2 − v1 ds = v1 − v2 ds. (5.535)
∂D2 ∂n ∂n ∂D1 ∂n ∂n
Next, we estimate the integral

I I
∂v1 1 ∂v1
v2 ds = h2 (ξ, η) − ln |(ξ, η) − (x2 , y2 )| ds, (5.536)
∂D2 ∂n ∂D2 2π ∂n
where we have used the definition of v2 in (5.530).

Observe that v1 in C 2 on Dεo (x2 , y2 ), since, in view of the definition of v1 in
(5.527), v1 is harmonic on Ω\{(x1 , y1 )}. Consequently, there exists a constant
M10 > 0 such that
|∇v1 (ξ, η)| 6 M10 , for all (ξ, η) ∈ Dεo (x2 , y2 ). (5.537)
Similarly, since h2 is harmonic in Ω, there exists a constant M2 > 0 such that
|h2 (ξ, η)| 6 M2 , for all (ξ, η) ∈ Dεo (x2 , y2 ). (5.538)
Consequently, using the estimates in (5.537) and (5.538), we obtain the estimate
I
∂v1
h2 (ξ, η) ds 6 M10 M2 2πε, for 0 < ε < εo . (5.539)
∂D2 ∂n

I
∂v1
lim+ h2 (ξ, η) ds = 0. (5.540)
ε→0 ∂Dε (x2 ,y2 ) ∂n
Next, use the estimate in (5.537) again to estimate

I
1 ∂v1
ln |(ξ, η) − (x2 , y2 )| ds 6 M10 |ε ln ε|, for 0 < ε < εo . (5.541)
∂D2 2π ∂n

I
1 ∂v1
lim+ ln |(ξ, η) − (x2 , y2 )| ds = 0. (5.542)
ε→0 ∂Dε (x2 ,y2 ) 2π ∂n
Combining the results in (5.540) and (5.542) with (5.536) we conclude that
I
∂v1
lim+ v2 ds = 0. (5.543)
ε→0 ∂Dε (x2 ,y2 ) ∂n
Similar arguments can be used to show that

I
∂v2
lim v1 ds = 0. (5.544)
ε→0 +
∂Dε (x1 ,y1 ) ∂n
Next, we evaluate the limit
I
∂v2
lim+ v1 ds. (5.545)
ε→0 ∂Dε (x2 ,y2 ) ∂n
Using (5.530) we can rewrite the integral in (5.545) as

I
∂h2 1 ∂
v1 − ln |(ξ, η) − (x2 , y2 ) ds;
∂Dε (x2 ,y2 ) ∂n 2π ∂n
so that,
I I I
∂v2 ∂h2 1 ∂
v1 ds = v1 ds− v1 ln |(ξ, η)−(x2 , y2 )| ds, (5.546)
∂D2 ∂n ∂D2 ∂n 2π ∂D2 ∂n
where we have written D2 for Dε (x2 , y2 ).
We estimate the first integral on the right–hand side of (5.546) as follows:
I
∂h2
v1 ds 6 M1 M20 2πε, for 0 < ε < εo , (5.547)
∂D2 ∂n
where
M1 = max |v1 (ξ, η)|,
(ξ,η)∈D εo (x2 ,y2 )
and
M20 = max |∇h2 (ξ, η)|.
(ξ,η)∈D εo (x2 ,y2 )

I
∂h2
lim+ v1 ds = 0. (5.548)
ε→0 ∂Dε (x2 ,y2 ) ∂n
Next, use the fact that
∂ d 1
ln |(ξ, η) − (x2 , y2 )| = ln r = , for all (ξ, η) ∈ ∂Dε (x2 , y2 ),
∂n dr r=ε ε
to evaluate the last expression on the right–hand side of (5.546) as follows:
I I
1 ∂ 1
v1 ln |(ξ, η) − (x2 , y2 )| ds = v1 ds, (5.549)
2π ∂D2 ∂n 2πε ∂Dε (x2 ,y2 )
the average value of v1 over the circle ∂Dε (x2 , y2 ). Thus, since v1 is continuous
on Dεo (x2 , y2 ), it follows from (5.549) that
I
1 ∂
lim+ v1 ln |(ξ, η) − (x2 , y2 )| ds = v1 (x2 , y2 ). (5.550)
ε→0 2π ∂D2 ∂n
Hence, letting ε → 0+ in (5.546), and using the limits in (5.548) and (5.550),
we get that I
∂v2
lim+ v1 ds = −v1 (x2 , y2 ). (5.551)
ε→0 ∂Dε (x2 ,y2 ) ∂n
Similar calculations to those leading to (5.551) can be used to show that
I
∂v1
lim+ v2 ds = −v2 (x1 , y1 ). (5.552)
ε→0 ∂Dε (x1 ,y1 ) ∂n
Finally, letting ε → 0+ in (5.535), and using the limits in (5.543), (5.544),

(5.551) and (5.552), we get that
v1 (x2 , y2 ) = v2 (x1 , y1 ),
which is the assertion in (5.532). We have therefore established the symmetry

of the Green’s function.
It follows from the definition of the Green’s function in (5.520)–(5.523), and
the fact that G((x, y), (ξ, η)), for (x, y) 6= (ξ, η), is symmetric, that the map
(x, y) 7→ G((x, y), (ξ, η)), for (x, y) 6= (ξ, η),
is harmonic as a function of (x, y), if (x, y) 6= (ξ, η). Using the definition of G
in (5.520), we then get that
H((x, y), (ξ, η)) = G((x, y), (ξ, η)) − W ((x, y), (ξ, η)),
for (x, y) 6= (ξ, η), is also symmetric. Consequently, the map
(x, y) 7→ H((x, y), (ξ, η)), for (x, y) ∈ Ω,
is harmonic as a function of (x, y) in Ω; so that,
Hxx + Hyy = 0, in Ω. (5.553)
Using the symmetry of G, we may define
G((x, y), (ξ, η)) = 0, for (x, y) ∈ ∂Ω and (ξ, η) ∈ Ω. (5.554)
Next, assume that f : Ω → R is Hölder continuous with Hölder α, with

0 < α 6 1. Then, according to Definition 5.4.1, there exists a constant M > 0
such that
|f (x, y) − f (ξ, η)| 6 M |(x, y) − (ξ, η)|α , for all (x, y), (ξ, η) ∈ Ω. (5.555)
It follows from (5.555) that f is continuous on Ω and that there exists a constant
Mo > 0 such that
|f (ξ, η)| 6 Mo , for all (ξ, η) ∈ Ω. (5.556)

(Recall that we are assuming that Ω is a bounded domain in R2 ).

Define u : Ω → R by
ZZ
u(x, y) = G((x, y), (ξ, η))f (ξ, η) dξdη, for (x, y) ∈ Ω. (5.557)
Ω
We will show that u solves Poisson’s equation
−∆u = f, in Ω.
Furthermore, it follows from the definition of u in (5.557) and from (5.554) that
u(x, y) = 0, for all (x, y) ∈ ∂Ω.
Thus, the function u given in (5.557) solves the Dirichlet problem for Poisson’s
equation given in (5.519); namely,

−∆u(x, y) = f (x, y), for (x, y) ∈ Ω;
(5.558)
u(x, y) = 0, for (x, y) ∈ ∂Ω.
Appendix A
Facts from the Theory of

Ordinary Differential
Equations
In this appendix we present some of the facts from the theory of ordinary dif-
ferential equations (ODEs) that are used in these notes. We begin with linear
second order ODEs with constant coefficients.
A.1 Linear, Second Order ODEs with Constant

Coefficients
We discuss here how to construct the general solution of the linear, homoge-
neous, second order ODE with constant coefficients
ay 00 + by 0 + cy = 0, (A.1)
where a, b and c are real constants, and y is assumed to be a twice differentiable

function of x ∈ R
We look for solutions of (A.1) of the form
y(x) = emx , for x ∈ R, (A.2)
where m is a parameter to be determined shortly.

Differentiating the function y defined in (A.2) and substituting into the ODE
in (A.1) yields the equation
am2 + bm + c = 0. (A.3)
The quadratic equation in (A.3) is called the characteristic equation of the

ODE in (A.1). Its roots could be real and distinct, real and equal, or complex
conjugates.
191
192 APPENDIX A. FACTS FROM THE THEORY OF ODES
If the roots, r1 and r2 , of the characteristic equation in (A.3) are real and
distinct, the general solution of the ODE in (A.1) is given by
y(x) = c1 er1 x + c2 er2 x , for all x ∈ R, (A.4)

If r ∈ R is the only root of the characteristic equation (A.3), the general
solution of the ODE in (A.1) is given by
y(x) = c1 erx + c2 xerx , for all x ∈ R, (A.5)

If the roots of the characteristic equation (A.3) are the complex conjugates
α ± iβ, the general solution of the ODE in (A.1) is given by
y(x) = c1 eαx cos βx + c2 eαx sin βx, for all x ∈ R, (A.6)

Appendix B
Theorems About
Integration
B.1 Differentiating Under the Integral Sign

Solutions of problems in the Calculus of Variations often require the differen-
tiation of functions defined in terms of integrals of other functions. In many
instance this involves differentiation under the integral sign. In this appendix
we preset a few results that specify conditions under which differentiation under
the integral sign is valid.
Proposition B.1.1 (Differentiation Under the Integral Sign). Suppose that

H : [a, b] × R → R is a C 1 function. Define h : R → R by
Z b
h(t) = H(x, t) dx, for all t ∈ R.
a
∂H
Assume that the functions H and are absolutely integrable over [a, b].
∂t
1
Then, h is C and its derivative is given by
Z b
0 ∂
h (t) = [H(x, t)] dx.
a ∂t
Proposition B.1.2 (Differentiation Under the Integral Sign and Fundamental

Theorem of Calculus). Suppose that H : [a, b] × R × R → R is a C 1 function.
Define Z t
h(y, t) = H(x, y, t) dx, for all y ∈ R, t ∈ R.
a
∂ ∂
Assume that the functions H, [H(x, y, t)] and [H(x, y, t)] are absolutely
∂y ∂t
193
194 APPENDIX B. THEOREMS ABOUT INTEGRATION
integrable over [a, b]. Then, h is C 1 and its partial derivatives are given by
Z t
∂ ∂
[h(y, t)] = [H(x, y, t)] dx
∂y a ∂y
and Z t
∂ ∂
[h(x, t)] = H(t, y, t) + [H(x, y, t)] dx.
∂t a ∂t
Proposition B.1.2 can be viewed as a generalization of the Fundamental

Theorem of Calculus and is a special case of Leibnitz Rule.
B.2 The Divergence Theorem

We begin by stating the two–dimensional version of the divergence theorem. We
then present some consequences of the result.
Let U denote an open subset of R2 and Ω a subset of U such that Ω ⊂ U .
We assume that Ω is bounded with boundary, ∂Ω, that can be parmetrized by
σ : [0, 1] → R, where σ(t) = (x(t), y(t)), for t ∈ [0, 1], with x, y ∈ C 1 ([0, 1], R)
satisfying
(ẋ(t))2 + (ẏ(t))2 6= 0, for all t ∈ [0, 1], (B.1)
(where the dot on top of the variable indicates derivative with respect to t), and
σ(0) = σ(1). Implicit in the definition of a parametrization is the assumption
that the map σ : [0, 1) → R2 is one–to–one on [0, 1). Thus, ∂Ω is a simple closed
curve in U . Observe that the assumption in (B.1) implies that at every point
σ(t) ∈ ∂Ω, a tangent vector
σ 0 (t) = (ẋ(t), ẏ(t)), for t ∈ [0, 1]. (B.2)

→
−
Let F : U → R2 denote a C 1 vector field in U ; so that,
→
−
F (x, y) = (P (x, y), Q(x, y)), for (x, y) ∈ U, (B.3)
where P : U → R and Q : U → R are C 1 , real–valued functions defined on U .

→
−
The divergence of the vector field F ∈ C 1 (U, R2 ) given in (B.3) is a scalar
→
−
field div F : U → R defined by
→
− ∂P ∂Q
div F (x, y) = (x, y) + (x, y) for (x, y) ∈ U. (B.4)
∂x ∂y
Example B.2.1. Imagine a two–dimensional fluid moving through a region U
in the xy–plane. Suppose the velocity of the fluid at a point (x, y) ∈ R2 is given
→
−
by a C 1 vector field V : U → R2 in units of distance per time. Suppose that we
also know the density of the fluid, ρ(x, y) at any point (x, y) ∈ U (in units of
mass per area), and that ρ : U → R is a C 1 scalar field. Define
→
− →
−
F (x, y) = ρ(x, y) V (x, y), for (x, y) ∈ U. (B.5)
B.2. THE DIVERGENCE THEOREM 195
→
− →
−
Then F has units of mass per unit length, per unit time. The vector field F in
(B.5) is called the flow field and it measures the amount of fluid per unit time
that goes through a cross section of unit length perpendicular to the direction of
→
−
V . Thus, to get a measure of the amount of fluid per unit time that crosses the
boundary ∂Ω in direction away from the region Ω, we compute the line integral
→
−
I
F ·nb ds, (B.6)
∂Ω
where ds is the element of arc–length along ∂Ω, and n b is unit vector that is
perpendicular to the curve ∂Ω and points away from Ω. The, expression in (B.6)
→
−
is called the flux of the flow field F across ∂Ω and it measures the amount of
fluid per unit time that crosses the boundary ∂Ω.
→
− →
−
On the other hand, the divergence, div F , of the flow field F in (B.5) has
units of mass/time × length2 , and it measures the amount of fluid that diverges
from a point per unit time per unit area. Thus, the integral
→
−
ZZ
div F dxdy (B.7)
Ω
the total amount of fluid leaving the reagin Ω per unit time. In the case where
there are not sinks or sources of fluid inside the region Ω, the integrals in (B.6)
and (B.7) must be equal; so that,
→
− →
−
ZZ I
div F dxdy = F ·nb ds. (B.8)
Ω ∂Ω
The expression in (B.8) is the Divergence Theorem.

Theorem B.2.2 (The Divergence Theorem in R2 ). Let U be an open subset
of R2 and Ω an open subset of U such that Ω ⊂ U . Suppose that Ω is bounded
with boundary ∂Ω. Assume that ∂Ω is a piece–wise C 1 , simple, closed curve.
→
−
Let F ∈ C 1 (U, R2 ). Then,
→
− →
−
ZZ I
div F dxdy = F ·nb ds, (B.9)
Ω ∂Ω
where n
b is the outward, unit, normal vector to ∂Ω that exists everywhere on
∂Ω, except possibly at finitely many points.
For the special case in which ∂Ω is parmatrized by σ ∈ C 1 ([0, 1], R2 ) satisfy-

ing (B.2), σ(0) = σ(1), the map σ : [0, 1) → R2 is one–to–one, and σ is oriented
in the counterclockwise sense, the outward unit normal to ∂Ω is given by
1
n
b(σ(t)) = (ẏ(t), −ẋ(t)), for t ∈ [0, 1]. (B.10)
|σ 0 (t)|
Note that the vector n b in (B.10) is a unit vector that is perpendicular to the
vector σ 0 (t) in (B.2) that is tangent to the curve at σ(t). In follows from (B.10)
196 APPENDIX B. THEOREMS ABOUT INTEGRATION
→
−
that, for the C 1 vector field F given in (B.3), the line integral on the right–hand
side of (B.9) can be written as
1
→
−
I Z
1
F ·n
b ds = (P (σ(t)), Q(σ(t))) · (ẏ(t), −ẋ(t)) |σ 0 (t)| dt,
∂Ω 0 |σ 0 (t)|
or
1
→
−
I Z
F ·n
b ds = [P (σ(t))ẏ(t) − Q(σ(t))ẋ(t)] dt,
∂Ω 0
which we can write, using differentials, as
→
−
I I
F ·nb ds = (P dy − Qdx). (B.11)
∂Ω ∂Ω
→
−
Thus, using the definition of the divergence of F in (B.4) and (B.11), we can
rewrite (B.9) as
ZZ I
∂P ∂Q
+ dxdy = (P dy − Qdx), (B.12)
Ω ∂x ∂y ∂Ω
which is another form of the Divergence Theorem in (B.9).

→
−
Applying the Divergence Theorem (B.9) to the vector field F = (Q, −P ),
where P, Q ∈ C 1 (U, R) yields from (B.12) that
ZZ I
∂Q ∂P
− dxdy = (P dx + Qdy),
Ω ∂x ∂y ∂Ω
which is Green’s Theorem.

As an application of the Divergence Theorem as stated in (B.12), consider
the case of the vector field (P, Q) = (x, y) for all (x, y) ∈ R2 . In this case (B.12)
yields ZZ I
2 dxdy = (xdy − ydx),
Ω ∂Ω
or I
2 area(Ω) = (xdy − ydx),
∂Ω
from which we get the formula
I
1
area(Ω) = (xdy − ydx), (B.13)
2 ∂Ω
for the area of the region Ω enclosed by a simple closed curve ∂Ω.
Appendix C
Kernels
C.1 The Dirichlet Kernel

In this section we derive the formula for the Dirichlet kernel (5.74) used in
connection with the proof of convergence of Fourier series.
The Dirichlet kernel was defined in (5.72) as
N
1 X nπθ
DN (θ) = + cos . for θ ∈ R. (C.1)
2 n=1 L
We will show that

1 πθ
sin N+
2 L
DN (θ) = , for θ 6= 0. (C.2)
πθ
2 sin
2L
We present two derivations of (C.2). The first one involves the use of Euler’s
formula
eiy = cos y + i sin y, for all y ∈ R. (C.3)
The second involves the use of some trigonometric identities.
πθ
Denote by x and compute
L
N
X 1 − ei(N +1)x
eix = , for x 6= 0. (C.4)
n=0
1 − eix
Observe that the real part of the expression on the left–hand side of (C.4) is
"N # N
X X
ix
Re e =1+ cos(nx), for x ∈ R, (C.5)
n=0 n=1
197
198 APPENDIX C. KERNELS
where we have used (C.3). We will compute the real part of
1 − ei(N +1)x (1 − ei(N +1)x )(1 − e−ix )

= , for x 6= 0, (C.6)
1 − eix (1 − eix )(1 − e−ix )
where
(1 − eix )(1 − e−ix ) = 2 − eix − e−ix = 2(1 − cos x), (C.7)
be virtue of Euler’s formula in (C.3).
Next, expand and simplify the numerator of the expression on the right–hand
side of (C.6) to get
(1 − cos[(N + 1)x] − i sin[(N + 1)x])(1 − cos x + i sin x),
or
1 − cos x − cos[(N + 1)x](1 − cos x) + sin[(N + 1)x]) sin x
+ i[(1 − cos[(N + 1)x]) sin x − sin[(N + 1)x](1 − cos x)].
It then follows that the real part of the expression on the right–hand side of
(C.4) is
1 − cos x − cos[(N + 1)x](1 − cos x) + sin[(N + 1)x]) sin x

,
2(1 − cos x)
or
1 − ei(N +1)x

1 cos[(N + 1)x] sin[(N + 1)x]) sin x
Re ix
= − + , (C.8)
1−e 2 2 2(1 − cos x)
for x 6= 0, where we have also used (C.7).

Next, use the trigonometric identities
x
1 − cos x = 2 sin2
2
and x x
sin x = 2 sin cos
2 2
to rewrite (C.8) as
x

1−e i(N +1)x

1 cos[(N + 1)x] sin[(N + 1)x]) cos
Re = − + x 2 ,
1 − eix 2 2 2 sin
2
which is turn can be rewritten as
x x

1−e i(N +1)x

1 sin[(N + 1)x]) cos − cos[(N + 1)x] sin
Re = + 2 2 ,
1 − eix 2 x
2 sin
2
C.1. THE DIRICHLET KERNEL 199
or
1
sin N + x
1 − ei(N +1)x

1 2
Re = + x , (C.9)
1 − eix 2 2 sin
2
where we have used the trigonometric identity
sin(A − B) = sin A cos B − cos A sin B, for A, B ∈ R.
In view of (C.4), (C.5) and (C.9), we can then write

1
N sin N + x
X 1 2
1+ cos(nx) = + x , for x 6= 0. (C.10)
n=1
2 2 sin
2
1
Finally, subtract from both sides of (C.10) to get
2

1
N sin N + x
1 X 2
+ cos(nx) = x , for x 6= 0,
2 n=1 2 sin
2
from which we derive (C.2), in view of (C.1).
Alternatively, use (C.1) to compute
X N
πθ 1 πθ πθ nπθ
sin DN (θ) = sin + sin cos , (C.11)
2L 2 2L n=1
2L L
and use the trigonometric identity
2 sin A cos B = sin(A + B) + sin(A − B)
to each of the terms in the sum in (C.11) to get
N X N
X πθ nπθ 1 1 πθ 1 1 πθ
sin cos = sin n + + sin −n ,
n=1
2L L n=1
2 2 L 2 2 L
or
N N
X πθ nπθ X 1 1 πθ 1 1 πθ
sin cos = sin n+ − sin n − ,
n=1
2L L n=1
2 2 L 2 2 L
since sin is an odd function. Note that the sum above telescopes to
N
X πθ nπθ 1 1 πθ 1 πθ
sin cos = sin N + − sin . (C.12)
n=1
2L L 2 2 L 2 2L
Combining (C.11) and (C.12) yields

πθ 1 1 πθ
sin DN (θ) = sin N + ,
2L 2 2 L
from which (C.2) follows.
200 APPENDIX C. KERNELS
Bibliography
[Ber83] H. C. Berg. Random Walks in Biology. Princeton University Press,

1983.
[CM93] A. J. Chorin and J. E. Marsden. A Mathematical Introduction to Fluid

Mechanics. Springer, 1993.
[LW55] M. J. Lighthill and G. B. Whitham. On kinematic waves ii: A theory
of traffic flow on long crowded roads. Proc. R. Soc. Lond., 229:317–345,
1955.
[Ric56] P. Richards. Shock waves on the highways. Operations Research,

4(1):42–51, 1956.
[Rud53] W. Rudin. Principles of Mathematical Analysis. McGraw–Hill, 1953.
[Tol62] G. P. Tolstov. Fourier Series. Dover Publications, 1962.
201

Math 180 Spring 2018 Notes

Uploaded by

Copyright:

Available Formats

Math 180 Spring 2018 Notes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Math 180 Spring 2018 Notes

Uploaded by

Copyright:

Available Formats

Notes on Partial Differential Equations

Preliminary Lecture Notes

2 Modeling with PDEs 7

5 Solving Linear PDEs 77

5.2 Fundamental Solutions . . . . . . . . . . . . . . . . . . . . . . . . 119

A Facts from the Theory of ODEs 191

B Theorems About Integration 193

This course is an introduction to the theory and applications of partial differ-

How Do PDEs Arise?

In general, a partial differential equation for a function, u, of several variables,

F (x, u, ux1 , . . . , uxn , ux1 x2 , . . . , uxn xn , . . .) = 0, (2.1)

where x = (x1 , . . . , xn ) and ux1 , . . . , uxn , ux1 x2 , . . . , uxn xn , . . . denote partial

for some constant k.

2.1 Modeling Fluid Flow

(x(t), y(t), z(t))

Figure 2.1.1: Region R

We assume throughout this discussion that ρ is a continuous function.

at (x, y, z) at time t = 0 will follow is determined by the following system of

2.1.1 The Continuity Equation

is given by equation (2.4),

Example 2.1.1 (Modeling Traffic Flow). Consider the unidirectional flow of

Figure 2.1.2: One–lane unidirectional flow

u = f (ρ, Λ), (2.22)

where f is a continuous function of ρ and a set of parameters, Λ. Some of the

Figure 2.1.3: Balance of Forces

Note that ZZZ

m(t) = mo , for all t, (2.28)

where mo is the constant given in (2.27). It follows from (2.28) that

In general, given a C 1 scalar field, g, defined in a region R, the material

Combining (2.29) with (2.33) we get that

It follows from (2.40) that

2.1.2 Conservation of Momentum for an Ideal Fluid

and which we’ll simply write as

where ~n is the outward unit normal to ∂Bt .

Applying the divergence theorem to each of these components we get

where ∇p is as given in (2.51).

according to the definition of momentum in (2.44).

which can be written as

which is an alternate form of the conservation of mass principle.

Substituting the expression for the rate of change of momentum in (2.60)

or any C 1 scalar field g,

2.1.3 Conservation of Energy in Incompressible Flow

It follows from (2.66) that

which we can rewrite as ZZZ

which can be written as

so that ZZZ ZZZ

2.1.4 Euler Equations for Incompressible, Ideal Fluids

stating the principles of conservation of mass, conservation of momentum, and

The fields ρ, ~u and p in (2.79) are assumed to be C 1 functions defined in and

~u · ~n = 0, on ∂R, for all t, (2.80)

2.2 Modeling Diffusion

Figure 2.2.4: Brownian Particles in a Region R

gives a measure of the number of particles that cross a unit cross–sectional at

where D is a proportionality constant known as the diffusion constant of the

u(x, 0) = f (x), for all x ∈ R,

Figure 2.2.5: Heat Conduction in a Cylindrical Rod

2.3 Variational Problems

2.3.1 Minimal Surfaces

Figure 2.3.6: Wire Loop