PDE Rumbos Math182Spring2014Notes

Notes on Partial Differential Equations
Preliminary Lecture Notes
Adolfo J. Rumbos

c Draft date April 22, 2014
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
3 Classification of PDEs 39
3.1 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2 Second Order PDEs . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Solving PDEs 47
4.1 Using Characteristic Curves . . . . . . . . . . . . . . . . . . . . . 47
4.1.1 Solving the One–Dimensional Wave Equation . . . . . . . 48
4.1.2 Solving First–Order PDEs . . . . . . . . . . . . . . . . . . 53
4.2 Using Symmetry to Solve PDEs . . . . . . . . . . . . . . . . . . . 60
4.2.1 Radially Symmetric Solutions . . . . . . . . . . . . . . . . 60
4.2.2 Dilation-Invariant Solutions . . . . . . . . . . . . . . . . . 65
4.2.3 Solving the Diffusion Equation . . . . . . . . . . . . . . . 69
5 Solving Linear PDEs 73

5.1 Fundamental Solutions . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.1 Heat Kernel . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.2 Uniqueness for the Diffusion Equation . . . . . . . . . . . 88
5.2 Dirichlet Problem for the Disk . . . . . . . . . . . . . . . . . . . 94
5.2.1 Separation of Variables . . . . . . . . . . . . . . . . . . . 96
5.2.2 An Eigenvalue Problem . . . . . . . . . . . . . . . . . . . 98
3
4 CONTENTS
5.2.3 Expansion in Terms of Eigenfunctions . . . . . . . . . . . 103

5.2.4 The Poisson Kernel . . . . . . . . . . . . . . . . . . . . . 110
5.2.5 The Poisson Integral . . . . . . . . . . . . . . . . . . . . . 116
A Differentiating Under the Integral Sign 123

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 derivatives in which we seek to find one of the functions, 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 differ-
entiability 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 exis-
tence of solutions. In a few cases, answering these 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. 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 nonex-
istence, 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.
The field of PDEs is vast and complex. The complexity is derived from
the great diversity of types of PDEs and of techniques for approaching their
analysis. PDEs range from linear to nonlinear; single equations to systems;
and from first degree to higher degree. There is also a further classification
determined by the behavior of solutions of certain classes of equations. Over
the years researchers have identified three major classes of PDEs: hyperbolic,
elliptic and parabolic. Archetypal instances of these classes of PDEs are the
classical equations of mathematical physics: the wave equation, Laplace’s or
Poisson’ equations, and the heat or diffusion equations, respectively. In these
notes we will provide examples of analysis for each of these types of equations.
Added to the complexity of the field of the PDEs is the fact that many problems
can be of mixed type. Hence, ability to recognize types of PDEs, or how a given
5
6 CHAPTER 1. PREFACE
problem can change from one type to the other, is very important in the analysis
of problems involving PDEs.
Finally, some classes of PDE problems have a particular structure that lends
itself to certain kind of general approach for analysis. For instance, many prob-
lems in PDEs can be can be formulated in terms of finding a function for which
a certain quantity is minimized, or maximized, over a class of functions. The
problem of finding such an optimizer is known as a variational problem. In these
notes we present an introduction to variational techniques for solving a class of
PDE problems that are amenable to the variational treatment.
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 (see Assignment #7). 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 (linear) density ρ(x) (in units
of mass per unit 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
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
string 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 form 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
2
1 L
Z
∂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
adding in length to L, and letting ∆x → 0, yields the expression in (2.127),
which we rewrite as
1 L 2
Z
K(t) = ρut 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 Lp
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 Lp Z L
1 L1 2
Z
2
1 2
1 + ux dx ≈ [1 + ux ] dx = L + u dx,
0 0 2 2 0 2 x
so that, substituting into (2.131),
1 L 2
Z
V (t) ≈ τ ux dx, for all t. (2.132)
2 0
Thus, in view of, we consider the problem of optimizing the quantity

Z T Z L
1 2 1 2
A(u) = ρut − τ ux dxdt, (2.133)
0 0 2 2
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) to the define a functional
in the class of functions A defines 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, (see Problem
1 in Assignment #8), 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, applying the result of Problem 2 in Assignment #6, we obtain from
(2.141) that
∂2u ∂2u
ρ 2 − τ 2 = 0, in R, (2.142)
∂t ∂x
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
2
= ,
∂t ρ ∂x2
or
∂2u 2
2∂ u
= c , (2.143)
∂t2 ∂x2
where
τ
c2 = ,
ρ
where the case in which ρ is assumed to be constant.
The wave equation in (2.142) or (2.143) is a second order, linear, hyperbolic
PDE.
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
39
40 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 41
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 43
(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 45
or
dt 1
=± ,
dx c
which we can rewrite as
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.
47
48 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 49
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 42. 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 to 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 61
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 R = {(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 R, (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 R to be the annulus
p
R = {(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 R;
u = a, on Cr1 ; (4.96)
u = b, on Cr2 ,

where a and b are real constants.

Solution: Since the annulus R 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) ∈ R, (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) ∈ R, (4.100)
ln(r2 /r1 ) ln(r2 /r1 )
to the BVP in (4.96). The result of Problem 3 in Assignment #7 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 42 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 presenting
some 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 Fundamental Solutions

We will illustrate the concept of a fundamental solution by first finding a special
solution to the one–dimensional diffusion equation.
5.1.1 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. (5.1)
∂t ∂x
73
74 CHAPTER 5. SOLVING LINEAR PDES
In Section 4.2.3 we derived the following dilation–invariant solution to the dif-

fusion equation in (5.1):
√
Z x/ t
2
u(x, t) = c1 e−z /4D
dz + c2 , for x ∈ R and t > 0, (5.2)
0
and constants c1 and c2 . Observe that the function u : R × (0, ∞) → R defined

in (5.2) 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 (5.1) 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 (5.1). Hence, by taking the partial derivative with respect to x in (5.2) we
obtain another solution to the the one–dimensional diffusion equation in (5.1).
Set v(x, t) = ux (x, t), for x ∈ R and t > 0, where u is given in (5.2). Then,
using the Fundamental Theorem of Calculus and the Chain Rule, we obtain
from (5.2) that
c1 2
v(x, t) = √ e−x /4Dt , for x ∈ R and t > 0, (5.3)
t
and some constant c1 , is a solution to the one–dimensional diffusion equation
in (5.1).
Z An interesting property of the function defined in (5.3) 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 −∞
5.1. FUNDAMENTAL SOLUTIONS 75
or Z ∞ √
v(x, t) dx = c1 4Dπ, for all t > 0. (5.4)
−∞
We chose the constant c1 in (5.4) so that

Z ∞
v(x, t) dx = 1, for all t > 0;
−∞
that is,
1
c1 = √ . (5.5)
4Dπ
Substituting the value of c1 in (5.5) into the definition of v(x, t) in (5.3), we
obtain
1 2
v(x, t) = √ e−x /4Dt , for x ∈ R and t > 0. (5.6)
4πDt
We shall denote the expression for v(x, t) defined in (5.6) by p(x, t), so that
1 2
p(x, t) = √ e−x /4Dt , for x ∈ R and t > 0. (5.7)
4πDt
It then follows from what we have shown thus far that the function p defined
in (5.7) is a C ∞ function defined in R × (0, ∞) that solves the one–dimensional
diffusion equation in (5.1); that is,
∂p ∂2p
= D 2, for x ∈ R and t > 0. (5.8)
∂t ∂x
Also, it follows from (5.4) and (5.5) that
Z ∞
p(x, t) dx = 1, for all t > 0. (5.9)
−∞
In fact, using a change of variables we obtain from (5.9) that

Z ∞
p(x − y, t) dy = 1, for all x ∈ R and t > 0. (5.10)
−∞
In addition to (5.10), the function p defined in (5.7) has the following properties:
Proposition 5.1.1 (Properties of p). Let p(x, t) be as defined in (5.7) 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 (5.10) and in Proposi-
tion 5.1.1 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 (5.11)


u(x, 0) = f (x), x ∈ R,
where f : R → R is a bounded function that is also piecewise continuous.

Definition 5.1.2 (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 5.1.1 shows a portion of the sketch of a piecewise continuous function.

We will show that the function u : R × (0, ∞) → R given by
Figure 5.1.1: Sketch of a Piecewise Continuous Function
Z ∞
u(x, t) = p(x − y, t)f (y) dy, for x ∈ R and t > 0, (5.12)
−∞
is a candidate for a solution of the initial value problem in (5.11). We note that,
since p(x − y, t)) is not defined at t = 0, the initial condition in the IVP in (5.11)
has to be understood as
lim u(x, t) = f (x).
t→0+
We will see in this section that (5.13) 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 5.1.3. Let u be given by (5.12), 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. (5.13)
∂t ∂x
Furthermore,
lim u(x, t) = f (x). (5.14)
t→0+
if f is continuous at x, and
f (x+ ) + f (x− )
lim+ u(x, t) = , (5.15)
t→0 2
if f has a jump discontinuity at x.
Once we have proved Proposition 5.1.3, 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 (5.7). Thus, a
solution of the initial value problem in (5.11) 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. (5.17)
4πDt
1 The function u is C 2 in the first variable, and C 1 in the second variable
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 5.1.4 (Properties of the Heat Kernel). Let K(x, y, t) be as defined

in (5.17) 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.

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 5.1.3, 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, (5.18)
π 0
and its properties:
Proposition 5.1.5. Let Erf : R → R be as given in (5.18). 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 5.1.2.
Lemma 5.1.6. Let p(x, t) be as defined in (5.7) for x ∈ R and t > 0. For δ > 0,
Z ∞
lim+ p(x, t) dx = 0. (5.19)
t→0 δ
and Z −δ
lim+ p(x, t) dx = 0. (5.20)
t→0 −∞
Figure 5.1.2: Sketch of Graph of Error Function
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 (5.18) 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. (5.21)
δ 2 4Dt
Now, it follows from (5.21) and (ii) in Proposition 5.1.5 that
Z ∞
lim p(x, t) dx = 0,
t→0+ δ
which is (5.19). Similar calculations can be used to derive (5.20).

Lemma 5.1.7. Let p(x, t) be as defined in (5.7) 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
(x − y, t) dy 6 1 , for all x ∈ R and t > 0,

∂t (5.22)
−∞
t
and
Z ∞
∂p

(x − y, t) dy = √ 1 ,

∂x for all x ∈ R and t > 0, (5.23)
−∞
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, (5.24)
4πDt
with respect to t to obtain
∂ 1 (x − y)2
[p(x − y, t)] = − p(x − y, t) + p(x − y, t), (5.25)
∂t 2t 4Dt2
for all x, y ∈ R and t > 0. Next, take absolute value on both sides of (5.25),
apply the triangle inequality, and use the positivity of the heat kernel (see (ii)
in Proposition 5.1.4) to get
2

[p(x − y, t)] 6 1 p(x − y, t) + (x − y) p(x − y, t),
∂
(5.26)
∂t 2t 4Dt2
for all x, y ∈ R and t > 0. Integrating on both sides of (5.26) and using (5.10)
(see (iii) in Proposition 5.1.4) yields
Z ∞ Z ∞
(x − y)2

∂
[p(x − y, t)] dy 6 1 +

∂t 2
p(x − y, t) dy, (5.27)
−∞
2t −∞ 4Dt
for all x ∈ R and t > 0.

Next, we evaluate the right–most integral in (5.27),
∞ ∞ 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
2
p(x − y, t) dy = √ ξ 2 e−ξ dξ, (5.28)
−∞ 4Dt t π −∞
for all x ∈ R and t > 0. The right–most integral in (5.28) can be evaluated
using integration by parts to yield
Z ∞ Z ∞
2 2
ξ 2 e−ξ dξ = 2 ξ 2 e−ξ dξ
−∞ 0
∞
2 ∞
Z
2
= −ξe−ξ + e−ξ dξ,
0 0
so that √
Z ∞
2 π
ξ 2 e−ξ dξ = . (5.29)
−∞ 2
Combining (5.29), (5.28) and (5.27) yields the estimate
Z ∞
∂
[p(x − y, t)] dy 6 1 , for x ∈ R and t > 0,

∂t t
−∞
which is (5.22).
In order to establish (5.23), first take the partial derivative with respect to
x on both side of (5.24) to get
∂ x−y
[p(x − y, t)] = − p(x − y, t), for x ∈ R and t > 0, (5.30)
∂x 2Dt
so that, taking absolute value on both sides of (5.30) and integrating,
Z ∞ Z ∞
∂
[p(x − y, t)] dy = |x − y|
∂x p(x − y, t) dy, (5.31)
−∞ 2Dt

−∞
Evaluate the right–most integral in (5.31),
∞ ∞ 2
|x − y| |x − y| e−(x−y) /4Dt
Z Z
p(x − y, t) dy = · √ dy, (5.32)
−∞ 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 = √ . (5.33)
−∞ 2Dt 4πDt πDt
The statement in (5.23) now follows by putting together the results in (5.33),
(5.32) and (5.31).
Proof of Proposition 5.1.3: Let f : R → R be a piecewise continuous function

satisfying
|f (x)| 6 M, for all x ∈ R, (5.34)
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 (5.24). We will show that
u solves the one–dimensional diffusion equation in (5.13). Before we do that,
though, we need to verify that the expression in (5.35) 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 (5.35) is a real number. This will follow from
the estimate
Z ∞
|p(x − y, t)f (y)| dy < ∞ for x ∈ R and t > 0. (5.36)
−∞
In order to derive the estimate in (5.36), use the positivity of the heat kernel
(see (ii) in Proposition 5.1.4), (5.10) and (5.34) 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, (5.37)
−∞
which implies (5.36). Observe that the estimate in (5.37) also implies that
|u(x, t)| 6 M, for x ∈ R and t > 0,
by virtue of (5.35).
The fact that u defined in (5.35) solves the one–dimensional diffusion equa-
tion in (5.13) 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; (5.38)
∂t ∂x
(see also (5.8). Indeed, suppose for the moment that we can interchange differ-
entiation and integration in the definition of u in (5.35), so that
Z ∞
∂u ∂p
(x, t) = (x − y, t)f (y) dy, for x ∈ R and t > 0, (5.39)
∂t −∞ ∂t
and
∞
∂2u ∂2p
Z
(x, t) = (x − y, t)f (y) dy, for x ∈ R and t > 0. (5.40)
∂x2 −∞ ∂x2
Thus, combining (5.39) and (5.40),

∞
∂2u ∂2p
Z
∂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 (5.13) holds true by virtue of (5.38)

The expressions in (5.39) and (5.40) are justified by the assumption that
f is bounded (see (5.34) and the estimates (5.22) and (5.23) in Lemma 5.1.7;
namely,
Z ∞
∂p
(x − y, t) dy 6 1 , for all x ∈ R and t > 0.

∂t t
−∞
and Z ∞

∂p

(x − y, t) dy = √ 1 ,

∂x for all x ∈ R and t > 0.
−∞
πDt
Observe that, (5.40) and (5.38) imply the estimate
Z ∞ 2
∂ p

1
∂x2 (x − y, t) dy 6 Dt , for all x ∈ R and t > 0.

−∞
We have therefore established that the function u : R × (0, 1) defined in (5.35) 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 5.1.3.

(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)| < . (5.41)
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 (5.35) and (5.10) (see also the
fact (iii) in Proposition 5.1.4). 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, (5.42)
−∞
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 (5.42) 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
−∞
(5.43)
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 (5.43), using
(5.41) and (5.10) to get
Z x+δ
ε
p(x − y, t)|f (y) − f (x)| dy < . (5.44)
x−δ 3
Next, use (5.34) and the triangle inequality to obtain the following estimate for
the last integral on the right–hand side of (5.43),
Z ∞ Z ∞
p(x − y, t)|f (y) − f (x)| dy 6 2M p(x − y, t) dy. (5.45)
x+δ x+δ
Make the change of variables ξ = y − x in the integral on the right–hand side

of (5.45) to obtain
Z ∞ Z ∞
p(x − y, t)|f (y) − f (xo )| dy 6 2M p(ξ, t) dξ, (5.46)
x+δ δ
where we have also used the symmetry of the heat kernel (see (i) in Proposition
5.1.4). It follows from (5.46) and (5.19) in Lemma 5.1.6 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 < . (5.47)
x+δ 3
Similar calculations to those leading to (5.47), using (5.20) in Lemma 5.1.6, 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 < . (5.48)
−∞ 3
Let δ3 = min{δ1 , δ2 }. It then follows from (5.43), in conjunction with (5.44),

(5.47) and (5.48), 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. (5.49)
t→0 −∞
It follows from (5.49) and the estimate in (5.42) that
lim |u(x, t) − f (x)| = 0,

t→0+
which yields (5.14) and assertion (i) of Proposition 5.1.3 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). (5.50)

y→x+ y→x−
Let ε > 0 be given. It follows from (5.50) that there exists δ > 0 such that
ε
x < y < x + δ ⇒ |f (y) − f (x+ )| < , (5.51)
3
and
ε
x − δ < y < x ⇒ |f (y) − f (x− )| < . (5.52)
3
Use the definition of u(x, t) in (5.35) to write
Z ∞
f (x+ ) + f (x− ) 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, (5.53)
2 −∞ xo
by virtue of (5.10), (5.9) and the symmetry of the heat kernel (see (i) in Propo-
sition 5.1.4). 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 (5.54)
−∞
Z ∞
+ p(x − y, t)|f (y) − f (x+ )| dy,
x
We re-write the last integral on the right–hand side of (5.54) 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 (5.55)
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 < , (5.56)
x 3 x 6
by virtue of (5.52) and (5.53).
Similar calculations to those leading to (5.47) 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 < . (5.57)
x+δ 3
Combining (5.56) and (5.57), we obtain from (5.55) that
Z ∞
ε
0 < t < δ1 ⇒ p(x − y, t)|f (y) − f (x+ )| dy < . (5.58)
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 < . (5.59)
−∞ 2
Thus, letting δ3 = min{δ1 , δ2 } we see that the conjunction of (5.58) and (5.59),
together with (5.54), implies that
f (x+ ) + f (x− )

0 < t < δ3 ⇒ u(x, t) − < ε.
2
We have therefore established (5.15) and the proof of part (ii) of Proposition
5.1.3 is now complete.
Example 5.1.8. Solve the initial value problem for the diffusion equation in
(5.11), where (
1, if − 1 < x 6 1;
f (x) = (5.60)
0, elsewhere.
Solution: A sketch of the graph of the initial condition, f , is shown in Figure
5.1.3. Note that f has jump discontinuities at −1 and at 1.
Figure 5.1.3: Initial Condition for Example 5.1.8
Using the formula in (5.35) we get that a solution to the initial value problem
(5.11) with initial condition given in (5.60) 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, (5.61)
4πDt −1
x−y
Make the change variables r = √ in (5.61) 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
−r 2 1 4Dt 2
u(x, t) = √ e dr − √ e−r dr, (5.62)
π 0 π 0
for x ∈ R and t > 0.

Making use of the error function defined in (5.18), we can rewrite (5.62) as

1 x+1 x−1
u(x, t) = Erf √ − Erf √ , (5.63)
2 4Dt 4Dt
for x ∈ R and t > 0. Figure 5.1.4 shows plots of the graph of y = u(x, t),
where u(x, t) is as given in (5.63), for various values of t in the case 4D = 1.
A few interesting properties of the function u given in (5.63) are apparent by
examining the pictures in Figure 5.1.4. First, the graph of y = u(x, t) is smooth
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 5.1.4: Sketch of Graph of y = u(x, t) for t = 0.1, 1, 10
for all t > 0. Even though the initial temperature distribution, f , in (5.60) is not
even continuous, the solution to the initial value problem (5.11) given in (5.63)
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 (5.63)) 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 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 5.1.4 that
lim u(x, t) = 0, for all x ∈ R. (5.64)

t→∞
5.1.2 Uniqueness for the Diffusion Equation

The observation (5.64) in Example 5.1.8 is true in general for solutions to the
initial value problem in (5.11) for the case in which the initial condition, f , is
square–integrable; that is,
Z ∞
|f (x)|2 dx < ∞ (5.65)
−∞
Observe that, for the function f in Example 5.1.8 satisfies

Z ∞
|f (x)|2 dx = 2,
−∞
so that the integrability condition in (5.65) holds true for the function in (5.60).
Before we establish that (5.64) is true for any solution of the initial value
problem (5.11) in which the initial condition satisfies (5.65), we will first need
to derive other properties of the function u given in (5.12).
Proposition 5.1.9. Let f : R → R be continuous and satisfying (5.65); 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. (5.66)
−∞
Then, Z ∞
|u(x, t)|2 dx < ∞, for all t > 0, (5.67)
−∞
and 2
Z ∞
∂u
(x, t) dx < ∞, for all t > 0. (5.68)
∂x
−∞
Proof: Let u be given by (5.66), where f satisfies the condition in (5.65). 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, (5.69)
−∞
where we have also used (5.10) and the positivity of the heat kernel (see (ii) and
(iii) in Proposition 5.1.4).
Integrate with respect to x on both sides of (5.69) to get
Z ∞ Z ∞Z ∞
|u(x, t)|2 dx 6 p(x − y, t)|f (y)|2 dydx, (5.70)
−∞ −∞ −∞
for t > 0. Interchanging the order of integration in the integral on the right–
hand side of (5.70) we obtain
Z ∞ Z ∞ Z ∞
|u(x, t)|2 dx 6 |f (y)|2 p(x − y, t) dx dy, (5.71)
−∞ −∞ −∞
for t > 0. It follows from (5.71) and (5.10) that

Z ∞ Z ∞
2
|u(x, t)| dx 6 |f (y)|2 dy, for t > 0, (5.72)
−∞ −∞
Combining (5.72) and (5.65) then yields

Z ∞
|u(x, t)|2 dx < ∞, for all t > 0, (5.73)
−∞
which is the condition in (5.67).

Next, differentiate u in (5.66) 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, (5.74)
∂x −∞ 2Dt

Proceeding as in the first part of this proof, use the Cauchy–Schwarz in-
equality (or Jensen’s inequality) to obtain from (5.74) that
2 Z ∞
(x − y)2

∂u
|f (y)|2 dy,

(x, t) 6 p(x − y, t) (5.75)
∂x
−∞ 4D2 t2

Next, integrate on both sides of (5.75) with respect to x and interchange the
order of integration to obtain
Z ∞ 2 Z ∞ Z ∞
∂u
(x, t) dx 6 1 2
(x − y)2 p(x − y, t) dx dy, (5.76)

|f (y)|
−∞ ∂x
4D2 t2 −∞ −∞
for t > 0.
Observe that the inner integral in the right–hand side of (5.76) 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. (5.77)
−∞
Putting together (5.76) and (5.77)
Z ∞
2 Z ∞
∂u
(x, t) dx 6 1 |f (y)|2 dy,

for t > 0,
−∞ ∂x 2Dt −∞

which implies (5.68) by virtue of (5.65).
We will next show that, if in addition to the integrability condition in (5.65)

for the initial distribution, f , we also impose the conditions (5.67) and (5.68) on
the initial value problem (5.11), then any solution must be of the form given in
(5.12). This amounts to showing that the initial value problem (5.11) in which
the initial condition satisfies (5.65), together with the integrability condition
in (5.67) and (5.68), has a unique solution. We will need the estimate in the
following lemma when we prove uniqueness.
Lemma 5.1.10. Let f : R → R be a continuous function satisfying (5.65). 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 ∞ (5.78)


 |u(x, t)|2 dx < ∞, for all t > 0;


 −∞


 2
 Z ∞

 ∂u
∂x (x, t) dx < ∞, for all t > 0.



−∞
Then, Z ∞ Z ∞
2
|v(x, t)| dx 6 |f (x)|2 dx, for t > 0. (5.79)
−∞ −∞
Proof: Let v denote any solution to the problem (5.78), where f satisfies the
integrability condition in (5.65).
In order to establish (5.79), set
Z ∞
E(t) = |v(x, t)|2 dx, for all t > 0. (5.80)
−∞
It follows from the integrability condition in (5.78) that E(t) in (5.80) 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, (5.81)
−∞
by virtue of the initial condition in problem (5.78).

Next, observe that, since v satisfies the diffusion equation in (5.78), that is
vt = Dvxx ,
then E is differentiable and

Z ∞ Z ∞
E 0 (t) = 2v(x, t)vt (x, t) dx = 2D v(x, t)vxx (x, t) dx, (5.82)
−∞ −∞
for t > 0.
We note that the integrability conditions in (5.78) imply that
lim v(x, t) = 0 and lim v(x, t) = 0, for t > 0, (5.83)

x→∞ x→−∞
and
lim vx (x, t) = 0 and lim vx (x, t) = 0. for t > 0, (5.84)
x→∞ x→−∞
Integrate by parts the last integral in (5.82) 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 2
Z ∞
∂v
0

E (t) = − ∂x (x, t) dx,
for t > 0, (5.85)
−∞
by virtue of (5.83), (5.84) and the last integrability condition in (5.78).

Now, it follows from (5.85) 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. (5.86)
The estimate in (5.79) follows from (5.86) in view of (5.80) and (5.81).
Proposition 5.1.11. Let f : R → R be a continuous function satisfying (5.65).
The problem
∂2u

∂u
= D 2, for x ∈ R, t > 0;



 ∂t ∂x





u(x, 0) = f (x), for x ∈ R;






Z ∞ (5.87)


 |u(x, t)|2 dx < ∞, for all t > 0;


 −∞



 Z ∞ 2

 ∂u
∂x (x, t) dx < ∞, for all t > 0,



−∞
has at most one solution.

Proof: Let v be any solution of the problem in (5.87) and let u be given by
(5.66). It follows from Proposition 5.1.3 and Proposition 5.1.9 that u solves
problem (5.87). Put
w(x, t) = v(x, t) − u(x, t), for x ∈ R and t > 0. (5.88)
It follows from the linearity of the differential equation in (5.87) that w also
solves the diffusion equation; indeed,
wt = vt − ut = Dvxx − Duxx = D(vxx − uxx ) = Dwxx .
The function w defined in (5.88) also satisfies the integrability condition in

problem (5.87) ; 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 , (5.89)
for all x ∈ R and all t > 0. Next, use the inequality
2ab 6 a2 + b2 , for a, b ∈ R,
in (5.89) to get
|w(x, t)|2 6 2 |v(x, t)|2 + |u(x, t)|2 ,

for x ∈ R and t > 0. (5.90)
Integrating on both sides of (5.90) 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 (5.87). Simi-
larly, 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
(5.87,
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 (5.78) in which the initial condition is the

constant function 0, it follows from the estimate (5.79) in Lemma 5.1.10 that
Z ∞
|w(x, t)|2 dx 6 0, for t > 0,
−∞

Z ∞
|w(x, t)|2 dx = 0, for t > 0. (5.91)
−∞
It follows from (5.91) 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 (5.88). Hence, any solution to the problem in
(5.87) must be that given by (5.66).
We will next show that, if u is any solution of problem (5.87), where f

satisfies the integrability condition
Z ∞
|f (x)|2 dx < ∞, (5.92)
−∞
then
lim u(x, t) = 0, for all x ∈ R. (5.93)
t→∞
To see why this is the case, apply Proposition 5.1.11 to write

Z ∞ −(x−y)2 /4Dt
e
u(x, t) = √ f (y) dy,
−∞ 4πDt
for all x ∈ R and t > 0, from which we get that
Z ∞ −(x−y)2 /4Dt
e
|u(x, t)| 6 √ |f (y)| dy, (5.94)
−∞ 4πDt
for all x ∈ R and t > 0. Next, square on both sides of (5.94) and apply the
Cauchy–Schwarz inequality to get
Z ∞ −(x−y)2 /2Dt Z ∞
1 e
|u(x, t)|2 6 √ √ dy |f (y)|2 dy, (5.95)
8πDt −∞ 2πDt −∞
where 2
∞
e−(x−y) /2Dt
Z
√ dy = 1. (5.96)
−∞ 2πDt
Combining (5.95) and (5.96), we then get
Z ∞
1
|u(x, t)|2 6 √ |f (y)|2 dy, (5.97)
8πDt −∞

It follows from (5.92) and (5.97) that
lim |u(x, t)|2 = 0, for all x ∈ R,

t→∞
which implies (5.93).
5.2 Solving the Dirichlet Problem in the Unit

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

uxx + uyy = 0 in D1 ;
(5.98)
u(x, y) = h(x, y), for (x, y) ∈ ∂D1 ,
5.2. DIRICHLET PROBLEM FOR THE DISK 95
∂D1
D1
x
Figure 5.2.5: Unit Disk in R2
where D1 = {(x, y) ∈ R2 | x2 + y 2 < 1} is the unit disk in R2 , and h is a given

function that is continuous in a neighborhood of the unit circle ∂D1 . 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, h, on the boundary of D1 .
The procedure that we will follow here construct a solution to the Dirichlet
problem in (5.2) illustrates an approach that has been successful in the solution
of boundary value problems for linear PDEs over domains with simple geometry.
• First, in view of the radial symmetry of the domain, we will express prob-
lem (5.2) in polar coordinate r and θ.
• 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 known as the method of separation of
variables.
• When looking solutions that are nonzero over the domain by means of
separation of variables, we are invariable lead 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
(5.98) and satisfies the boundary condition in that problem; this method
is usually referred to as eigenfunctions expansion.
5.2.1 Separation of Variables

In view of the radial symmetry of the domain (see Figure 5.2.5), 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
(5.98) and use the principle of superposition to construct a solution of the prob-
lem 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 (5.98) in polar coordinates:
 2
∂ v 1 ∂v 1 ∂2v

 + + = 0, r > 0, −π < θ 6 π;
∂r2 r ∂r r2 ∂θ2 (5.99)


v(1, θ) = h(cos θ, sin θ), −π < θ 6 π,
where we have set

v(r, θ) = u(r cos θ, r sin θ).
(See Problem 3 in Assignment #11). We will denote h(cos θ, sin θ) by g(θ),
where g : R → R is a continuous periodic function of period 2π. We can then
rewrite the BVP in (5.99) as
 2
∂ v 1 ∂v 1 ∂2v

 2
+ + 2 2 = 0, r > 0, −π < θ 6 π;
∂r r ∂r r ∂θ (5.100)


v(1, θ) = g(θ), −π < θ 6 π.
We start out by looking for special solutions of the PDE in (5.100) of the form
v(r, θ) = f (r)z(θ), for r > 0 and − π < θ < π, (5.101)
where f : [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) = f 0 (r)z(θ), r > 0, −π < θ 6 π;
∂r
∂2v
(r, θ) = f 00 (r)z(θ), r > 0, −π < θ 6 π;
∂r2
∂2v
(r, θ) = f (r)z 00 (θ), r > 0, −π < θ 6 π,
∂θ2
and substitute them into the PDE in (5.100) to obtain
1 1
f 00 (r)z(θ) + f 0 (r)z(θ) + 2 f (r)z 00 (θ) = 0, for r > 0, −π < θ 6 π. (5.102)
r r
Assuming that v(r, θ) is not zero for all values of r and θ, and dividing on both
sides of (5.102) by v(r, θ) as given in (5.101), we obtain
f 00 (r) 1 f 0 (r) 1 z 00 (θ)

+ + 2 = 0, for r > 0, −π < θ 6 π. (5.103)
f (r) r f (r) r z(θ)
Multiplying on both sides of the equation in (5.103) 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:
f 00 (r) f 0 (r) z 00 (θ)

r2 +r =− , for r > 0, −π < θ 6 π. (5.104)
f (r) f (r) z(θ)
Since (5.104) holds true for all values of r and θ in (0, 1) and (−π, π], respectively,
it follows from (5.104) that each side of the equation in (5.104) must be equal
to a constant.2 Call that constant λ so that
f 00 (r) f 0 (r) z 00 (θ)
r2 +r =− = λ, for r > 0, −π < θ 6 π. (5.105)
f (r) f (r) z(θ)
The expression in (5.105) leads to two ordinary differential equations
−z 00 (θ) = λz(θ), for − π < θ 6 π, (5.106)
and
r2 f 00 (r) + rf 0 (r) = λf (r), for r > 0. (5.107)
The requirement that the function g in (5.100) be periodic of period 2π yields
the following conditions for z:
z(−π) = z(π) and z 0 (−π) = z 0 (π); (5.108)
in other words, we will assume that z can be extended to a C 1 periodic function

defined on R with period two 2π. Putting together (5.106) and (5.108) yields
the following two–point boundary value problem:
 −z 00 (θ) = λz(θ),

for − π < θ < π;
z(−π) = z(π); (5.109)
 0
z (−π) = z 0 (π).
2 To see why this assertion is true, pick θo in (−π, π] such that z(θo ) 6= 0; then, by virtue
f 00 (r) f 0 (r) z 00 (θo )
of (5.104), r2 +r =− , for all r > 0; so that the left–hand side of (5.104)
f (r) f (r) z(θo )
z 00 (θ)
is constant. Similarly, for fixed ro in (0, 1) with f (ro ) 6= 0, (5.104) implies that =
z(θ)
f 00 (r ) f 0 (r )
o o
−ro2 − ro , for all θ in (−π, π], so that the right–hand side of (5.104) must also
f (ro ) f (ro )
be constant.
5.2.2 An Eigenvalue Problem

Observe that the constant function z(θ) = 0, for all values of θ, solves two–point
BVP in (5.109); we shall refer to this solution as the trivial solution. We are
interested in nontrivial solutions of (5.109); otherwise, the special solutions in
(5.101) of the BVP in (5.100) 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 (5.100) for arbitrary (nonzero) boundary conditions. We will see shortly that
the answer to the question of whether or not the two–point BVP in (5.109)
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 (5.109) has nontrivial
solutions; for the rest of the values of λ the two–point BVP (5.109) has only the
trivial solution.
Definition 5.2.1 (Eigenvalues and Eigenfunctions). A value of λ in (5.109) for

which the two–point BVP in (5.109) 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 (5.109). Before we proceed with the calculations, it will be helpful to
know that the eigenvalues of (5.109) must be nonnegative. We state that fact
in the following proposition.
Proposition 5.2.2. Assume that the two–point BVP (5.109) has nontrivial
solution. Then, λ > 0.
Proof: Let z be a nontrivial solution of (5.109). Multiply the ODE in (5.109)

by z and integrate from −π to π to get
Z π Z π
00
− z (θ)z(θ) dθ = λ z(θ)z(θ) dθ. (5.110)
−π −π
Use integration by parts to evaluate the left–most integral in (5.110) 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 (5.109),

Z π Z π
00
z (θ)z(θ) dθ = − [z 0 (θ)]2 dθ. (5.111)
−π −π
Substituting the result in (5.111) into the left–hand side of (5.110) then yields
Z π Z π
[z 0 (θ)]2 dθ = λ [z(θ)]2 dθ. (5.112)
−π −π
Since z is a nontrivial solution of the two–point BVP in (5.109), it follows that

Z π
[z(θ)]2 dθ > 0. We can therefore solve (5.112) 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 5.2.2, it suffices to look for nontrivial
solutions of (5.109) for either λ = 0 or λ > 0.
For the case in which λ = 0 in (5.109), the ODE in (5.109) becomes
z 00 (θ) = 0,
which has general solution

z(θ) = c1 θ + c2 , (5.113)
for arbitrary constants c1 and c2 .
Applying the first boundary condition to z given in (5.113) yields
−πc1 + c2 = πc2 + c2 ,
from which we get that 2πc1 = 0, so that c1 = 0. It then follows from (5.113)
any solution of the BVP in (5.109) with λ = 0 must be constant:
z(θ) = c, for all θ. (5.114)
In particular, if c 6= 0 in (5.114), z(θ) = c for all θ is a nontrivial solution of the

two–point BVP (5.109). Consequently, λ = 0 is an eigenvalue of (5.109). For
future reference, we shall denote this eigenvalue by λo , so that
λo = 0, (5.115)
and we shall pick the special eigenfunction
ϕo (θ) = 1, for all θ, (5.116)
and note that any solution of the BVP in (5.109) for λo is a constant multiple
of ϕo given in (5.116); so that
zo (θ) = ao , for all θ, (5.117)
where ao denotes a real constant, represents all solutions of the two–point BVP
in (5.116) corresponding to to the eigenvalue λo = 0.
Next, we look for positive eigenvalues of the BVP in (5.109). For the case
in which λ > 0 in (5.109), the general solution of the ODE in (5.109) is
√ √
z(θ) = c1 cos( λθ) + c2 sin( λθ), for all θ, (5.118)
and arbitrary constants c1 and c2 , so that

√ √ √ √
z 0 (θ) = −c1 λ sin( λθ) + c2 λ cos( λθ), for all θ. (5.119)
Imposing the the boundary conditions in (5.109) to the functions given in (5.118)
and (5.119) 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);
√ (5.120)
thus, dividing the second equation in (5.120) 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;
(5.121)
2c1 sin( λπ) = 0.
Since we are looking for nontrivial solutions of (5.109), we require that c1 and
c2 in (5.118) are not both zero. Consequently, we obtain from (5.121) that
√
sin( λπ) = 0. (5.122)
Solutions to the trigonometric equation in (5.122) are given by

√
λπ = nπ (5.123)
where n is an integer. It follows from (5.123) that the positive eigenvalues of

the BVP in (5.109) are given by
λ = n2 , for n = 1, 2, 3, . . . . (5.124)
We will denote the positive eigenvalues of the BVP (5.109) in (5.124) by λn , for
n = 1, 2, 3, . . ., so that
λn = n2 , for n = 1, 2, 3, . . . . (5.125)
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, (5.126)
where an and bn , for n = 1, 2, 3 . . ., are real constants.

We shall put together the results in (5.115), (5.117), (5.125) and (5.126) in
the following proposition:
Proposition 5.2.3 (Eigenvalues and eigenfunctions of BVP (5.109)). The eigen-

values of the two–point BVP (5.109) are given by
λn = n2 , for n = 0, 1, 2, 3, . . . , (5.127)
with corresponding eigenfunctions of the form
zo (θ) = ao , for all θ ∈ R,
and
zn (θ) = an cos(nθ) + bn sin(nθ) for n = 1, 2, 3, . . . , and θ ∈ R,
where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are real constants.

We the values for λ given in (5.127) we now proceed to solve the ODE in
(5.107) for the radial component of the special solutions to the BVP in (5.100)
of the form given in (5.101); namely,
r2 f 00 (r) + rf 0 (r) = n2 f (r), for r > 0 and n = 0, 1, 2, . . . (5.128)
We shall first solve (5.128) for the case n = 0. In this case the equation becomes
rf 00 (r) + f 0 (r) = 0, for r > 0, (5.129)
where we have divided by r > 0. Observe that the equation in (5.129) can be
written as
d
[rf 0 (r)] = 0, for r > 0,
dr
which can be integrated to yield
rf 0 (r) = c1 , for r > 0,
and some constant c1 , or

c1
f 0 (r) = , for r > 0, (5.130)
r
and some constant c1 . Integrating the equation in (5.130) then yields
f (r) = c1 ln(r) + c2 , for r > 0, (5.131)
and some constants c1 and c2 . Observe that, if c1 6= 0 in (5.131), the the

function f given a (5.131) is not unbounded as r → 0+ . Thus, since we are
looking for C 2 functions defined in the closure of the unit disk, D1 , we must set
c1 equal to 0. This is equivalent to imposing the following boundary condition
on f :
lim f (r) exists. (5.132)
r→0+
Hence, it follows from (5.131) and (5.132) that, for n = 0, a solution for (5.128)
is given by
f (r) = c, for all r, (5.133)
is a solution, for some constant c. Taking c = 1 in (5.133) we get the solution

to (5.128) corresponding to n = 0:
fo (r) = 1, for all r. (5.134)
Next, consider the case n > 1 in (5.128). In this case the differential equation
in (5.128) is an ODE of Euler type:
r2 f 00 (r) + rf 0 (r) − n2 f (r) = 0, for r > 0. (5.135)
The ODE in (5.135) can be solved by looking for solutions of the form
f (r) = rq , for r > 0 (5.136)
and some real number q.

Taking derivatives of f in (5.136) and substituting into (5.136) 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. (5.137)
q(q − 1) + q − n2 = 0,
or
q 2 − n2 = 0,
or
(q + n)(q − n) = 0,
q = ±n. (5.138)
It follows from (5.137) and (5.138) that
f−n (r) = r−n and fn (r) = rn , for r > 0. (5.139)
In view of the boundary condition in (5.132), we take the second solution in

(5.139),
fn (r) = rn , for all r and n = 1, 2, 3, . . . . (5.140)
Putting together (5.140), (5.134), (5.126), (5.117), and (5.101), we conclude
that we have found an infinite collection of solutions to the PDE in (5.100);
namely,
vo (r, θ) = ao , for all r and θ; (5.141)
vn (r, θ) = rn [an cos(nθ) + bn sin(nθ)], for all r and θ, (5.142)
where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are real constants.
5.2.3 Expansion in Terms of Eigenfunctions

None of the functions in (5.141) and (5.142) by itself will satisfy the general
boundary condition in (5.100). We can, however, attempt to construct a solution
to (5.100) 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. (5.143)
n=0
provided the series in (5.143) converges to a C 2 function.

Let’s assume for the moment that the series in (5.143) converges also for
r = 1, so that we can apply the boundary condition in (5.100) to get
∞
X
ao + [an cos(nθ) + bn sin(nθ)] = g(θ), for − π < θ 6 π. (5.144)
n=0
Assuming for the moment that the series on the left–hand side of (5.144) 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 g by means of the following integration facts:
Z π
sin(nθ) cos(mθ) dθ = 0, for all m, n = 1, 2, 3, . . . ; (5.145)
−π
(
π
if m 6= n;
Z
0,
cos(nθ) cos(mθ) dθ = (5.146)
−π π, if m = n;
and (
π
if m 6= n;
Z
0,
sin(nθ) sin(mθ) dθ = (5.147)
−π π, if m = n.
Indeed, integrating on both sides of (5.144) from −π to π we get, assuming that
the series in (5.144) can be integrated term by term,
Z π
2πao = g(θ) dθ,
−π
from which we get Z π

1
ao = g(θ) dθ; (5.148)
2π −π
thus, ao is the average value of g over the interval (−π, π].

Next, multiply the equation in (5.144) on both sides by cos(mθ) to obtain
∞
X
ao cos mθ + [an cos nθ cos mθ + bn sin nθ cos mθ] = g(θ) cos mθ. (5.149)
n=0
Then, integrate on both sides of (5.149) with respect to θ from −π to π, and

use the identities in (5.145) and (5.146) to get
Z π
πam = g(θ) cos(mθ) dθ,
−π
from which we get

Z π
1
am = g(θ) cos(mθ) dθ, for m = 1, 2, 3, . . . . (5.150)
π −π
Similar calculations (this time multiplying the equation in (5.144) on both sides
by sin(mθ), integrating from −π to π, and using the integral identities in (5.145)
and (5.147)) lead to
1 π
Z
bm = g(θ) sin(mθ) dθ, for m = 1, 2, 3, . . . . (5.151)
π −π
The numbers defined in (5.148), (5.150) and (5.151) are called the Fourier
coefficients of the 2π–periodic function g. Note that the Fourier coefficients of
g are defined whenever g is absolutely integrable over the interval [−π, π].
Definition 5.2.4 (Absolute Integrability). A function g : [−π, π] → R is said
to be absolutely integrable over [−π, π] whenever
Z π
|g(θ)| dθ < ∞. (5.152)
−π
Note that g doesn’t have to be continuous for (5.152) for (5.152). For in-
stance, if g is bounded and piece–wise continuous then (5.152) holds; indeed,
suppose that g piece–wise continuous and
|g(θ)| 6 M, for θ ∈ [−π, π],
and some positive constant M ; then

Z π Z π
|g(θ)| dθ 6 M dθ = 2πM < ∞.
−π −π
Notation 5.2.5. We will denote the integral in (5.152) by kgkL1 ; so that

Z π
kgkL1 = |g(θ)| dθ. (5.153)
−π
If the integral in (5.153) ia understood as the Lebesgue integral, and kgkL1 < ∞
we will say that g is an L1 function and write g ∈ L1 (−π, π). We shall refer to
kgkL1 as the L1 norm of g ∈ L1 (−π, π).
The existence of the Fourier coefficients of g in (5.148), (5.150) and (5.151)
is guaranteed for absolutely integrable 2π–periodic functions, g, or for g ∈
L1 (−π, π). This is the content of the following proposition.
Proposition 5.2.6 (Existence of the Fourier Coefficients). Let an , for n =

0, 1, 2, . . ., be as given in (5.148) and (5.150), and bn , for n = 1, 2, 3, . . ., be as
in (5.151), where g ∈ L1 (−π, π). Then,
1
|an | 6 kgkL1 , for n = 0, 1, 2, 3, . . . ; (5.154)
π
and
1
|bn | 6 kgkL1 , for n = 1, 2, 3, . . . . (5.155)
π
Proof: The estimates in (5.154) and (5.155) follow from properties of the inte-
gral. For ao , we get from (5.148) that
Z π
1
|ao | 6 |g(θ)| dθ,
2π −π
so that, using the definition of the L1 norm of g in (5.153),
1 1
|ao | 6 kgkL1 6 kgkL1 .
2π π
For n = 1, 2, 3, . . . we obtain from (5.150) that
1 π
Z
|an | 6 |g(θ)| | cos(nθ)| dθ
π −π
Z π
1
6 |g(θ)| dθ,
π −π
since | cos(nθ)| 6 1 for all θ and all n, which yields (5.154). Similar calculations
lead to (5.155).
It follows from Proposition 5.2.6 that the sequences of Fourier coefficients,

(an ) and (bn ), of g are bounded by a constant depending on the L1 norm of
g. 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 5.2.7 (Riemann–Lebesgue Lemma). Let an , for n = 0, 1, 2, . . .,

be as given in (5.148) and (5.150), and bn , for n = 1, 2, 3, . . ., be as in (5.151),
where g ∈ 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 5.2.6 to prove that the series
defining the function v in (5.143) converges in D1 .
Proposition 5.2.8 (Point–wise Convergence of Series in (5.143) ). Let an , for

n = 0, 1, 2, . . ., be as given in (5.148) and (5.150), and bn , for n = 1, 2, 3, . . .
be as in (5.151), where g ∈ L1 (−π, π). Then, the series defining v in (5.143)
converges absolutely in D1 .
Proof: The conclusion will follow by comparing with the geometric series since
0 6 r < 1 and
kgkL1 n
|rn an cos(nθ)| 6 rn |an | 6 r , for all n,
π
where we have used the estimate (5.154) in Proposition 5.2.6. Similarly, using
(5.155) in Proposition 5.2.6,
kgkL1 n
|rn bn sin(nθ)| 6 r ,
π
for all n.
Proposition 5.143) allows us to conclude that the function v given in (5.143)
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
(5.143), 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 g.
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 (5.144) converges
uniformly. Uniform convergence will justify the term–by–term integration that
was done in order to obtain the formulas in (5.148), (5.150) and (5.151). We will
denote the trigonometric series on the left–hand side of the equation in (5.144)
by gb(θ), so that
∞
X
gb(θ) = ao + [an cos(nθ) + bn sin(nθ)], for − π 6 θ 6 π, (5.156)
n=0
where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are the Fourier coef-

ficients of g. Let (b
gn (θ)) denote the sequence of partial sums of the series in
(5.156) so that
k
X
gbn (θ) = ao + [ak cos(kθ) + bk sin(kθ)], for − π 6 θ 6 π. (5.157)
k=0
Definition 5.2.9 (Uniform Convergence). We say that sequence of functions,

(b
gn ), defined in (5.157) converges uniformly to g in [−π, π] if
lim max |b
gn (θ) − g(θ)| = 0.
n→∞ −π6θ6π
The following proposition gives a sufficient condition for the trigonometric

series in (5.156) to converge uniformly to g.
Theorem 5.2.10 (Uniform Convergence). Assume that g : R → R is a contin-

uous 2π–periodic function; assume also that g is piece–wise differentiable with
g 0 : R → R piecewise continuous. Let (bgn ) be the sequence of trigonometric func-
tions defined in (5.156), where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . .,
are the Fourier coefficients of g. Then, (bgn ) converges uniformly to g in [−π, π]
as n → ∞.
A proof of Theorem 5.2.10 may be found in [Tol62, pp. 80-81]. The idea of
the proof is to derive the estimate
∞
X
(|ak | + |bk |) < ∞, (5.158)
k=0
for the Fourier coefficients of a piece–wise C 1 , 2π–periodic function g. The

uniform convergence of the series in (5.156) then follows from the observation
that sum of the absolute values of the terms in the series on the right–hand side
of (5.156) is bounded above by the series on the left–hand side of (5.158). The
uniform convergence of the series in (5.156) is then a consequence of Weierstrass
Majorization Test, or Weierstrass M–Test for uniform convergence, (see, for
example, [Rud53, Theorem 7.10, pg. 119]).
Definition 5.2.11 (Piece–wise C 1 ). We say that g : R → R is piecewise C 1 if

it is differentiable, and its derivative is piece–wise continuous.
Let’s assume for the moment that g is a 2π–periodic, piece–wise C 1 function.

It then follows from Theorem 5.2.10 that the Fourier series on the left–hand
side of (5.144), where an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are
the Fourier coefficients of g, 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 (5.148), (5.150) and (5.151) by virtue
of the following theorem form Analysis:
Theorem 5.2.12 (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 5.2.8 that the trigonometric series defining v(r, θ) in
(5.143),
∞
X
v(r, θ) = ao + rn [an cos nθ + bn sin nθ], 0 6 r 6 1, −π 6 θ 6 π, (5.159)
n=0
converges in D1 , provided that g ∈ 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 g is piece–wise C 1 and 2π–periodic, then the series in
(5.159), where the an , for n = 0, 1, 2, . . ., and bn , for n = 1, 2, 3, . . ., are the
Fourier coefficients of g, converges uniformly in D1 , the closed unit disk in R2 .
Proposition 5.2.13 (Uniform Convergence of Series in (5.143) ). Let an , for
n = 0, 1, 2, . . ., be as given in (5.148) and (5.150), and bn , for n = 1, 2, 3, . . . be
as in (5.151), where g is a piecewise C 1 , 2π–periodic function. Then, the series
defining v in (5.159) converges uniformly in D1 .
Proof: The assumptions that g is piece–wise C 1 and 2π periodic imply that the
Fourier coefficients of g satisfy the estimate in (5.158); namely,
∞
X
(|ak | + |bk |) < ∞. (5.160)
k=0
See [Tol62, pp. 80-81] for details of the calculations leading up to (5.160).
Next, us the triangle inequality to estimate the absolute values of the terms
of the series in (5.159) 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 (5.159) are “majorized” by the terms of the convergent series in
(5.160). It then follows by the the Weierstrass M–Test for uniform convergence
([Rud53, Theorem 7.10, pg. 119]) that the series in (5.159) converges uniformly
for r ∈ [0, 1] and θ ∈ [−π, π].
We will get a chance to use Weierstrass M–Test for uniform convergence
once again to justify the following calculations based on the trigonometric series
representation for v(r, θ) in (5.159) and the assumption that g is a piecewise C 1 ,
2π–periodic function.
First, substitute the formulas defining the Fourier coefficients of g in (5.148),
(5.150) and (5.151) into the right–hand side (5.159) to get
Z π ∞ Z π
1 X
n 1
v(r, θ) = g(ξ) dξ + r g(ξ) cos(nξ) dξ cos(nθ)
2π −π n=0
π −π
Z π
1
+ g(ξ) sin(nξ) dξ sin(nθ)
π −π
Z π ∞ Z π
1 1X n
= g(ξ) dξ + r cos(nθ) cos(nξ)g(ξ) dξ
2π −π π n=0 −π
Z π
+ sin(nθ) sin(nξ)g(ξ) dξ
−π
Z π
1
= g(ξ) dξ
2π −π
∞ Z
1X π n
+ r [cos(nθ) cos(nξ) + sin(nθ) sin(nξ)]g(ξ) dξ,
π n=0 −π

Z π ∞ Z
1 1X π n
v(r, θ) = g(ξ) dξ + r cos[n(θ − ξ)]g(ξ) dξ, (5.161)
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 (5.161).

This is justified by the fact that the series
∞
X
rn cos(nξ)
n=1
converges 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 (5.161), we
can write
Z π " ∞
#
1 X
n
v(r, θ) = 1+ 2r cos[n(θ − ξ)] g(ξ) dξ, (5.162)
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], (5.163)
2π n=1
we see that (5.164) can be written as

Z π
v(r, θ) = P (r, θ − ξ)g(ξ) dξ, for 0 6 r < 1 and θ ∈ [−π, π]. (5.164)
−π
5.2.4 The Poisson Kernel for the Unit Disk

The function P defined in (5.163) is called the Poisson kernel for the unit
disk in R2 , and the expression on the right–hand side of (5.164) 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 (5.164).
We will first show that the series defining the Poisson kernel in (5.163)
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 estimates the absolute values of the of the
terms of the series,
|2rn cos(nθ)| 6 2rn , for all θ ∈ [−π, π]. (5.165)
It follows from (5.165) that the absolute values of the terms of the series in
(5.163) 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 (5.163) 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 (5.163) 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 5.2.14 (Uniform Limit of Continuous Functions). Let (fn ) be a

sequence of continuous functions on [a, b] that converges uniformly to a functions
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) (5.166)
n=1 n=1
converge uniformly. This assertion will follow from the following proposition
Proposition 5.2.15 (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 uniformly over [a, b]. Assume also that the series
∞
X
uk (xo )
k=1
converges at some point xo in [a, b]. Then, the series converges

∞
X
uk
k=1
converges uniformly to function f that is differentiable over [a, b], and

∞
X
0
f (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 (5.166) 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 (5.166) are “majorized” by the series
∞
X ∞
X
2nrn−1 and 2nrn , (5.167)
n=1 n=1
respectively; both of the series in (5.167) converge by the Root Test (or the
Ratio Test), since 0 6< 1. It then follows by the Weierstrass M–Test for
uniform convergence, that the series in (5.167) converge uniformly in θ. The
same argument applied to r ∈ [0, R], where where R < 1, yields that the series
in (5.166) are absolutely and uniformly convergent for θı[−π, π] and r ∈ [0, R].
This time the series in (5.166) are “majorized” by the convergent series
∞
X ∞
X
2nRn−1 and 2nRn .
n=1 n=1
It the follows from Proposition 5.2.15 that the partial derivatives of the Poisson
kernel in (5.163) have partial derivatives in D1 given by
∞
∂ 1 X
[P (r, θ)] = 2nrn−1 cos(nθ), 0 6 r < 1 and θ ∈ [−π, π], (5.168)
∂r 2π n=1
and
∞
∂ 1 X
[P (r, θ)] = − 2nrn sin(nθ), 0 6 r < 1 and θ ∈ [−π, π], (5.169)
∂θ 2π n=1
where we have differentiated the series in (5.163) term–by–term. A similar

argument can be used to obtain the second partial derivatives of the of the
Poisson kernel:
∞
∂2 1 X
[P (r, θ)] = 2n(n − 1)rn−2 cos(nθ), 0 6 r < 1, θ ∈ [−π, π], (5.170)
∂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], (5.171)
∂θ2 2π n=1
where the series in (5.168) and (5.169) have been differentiated term–by–term.
Next, substitute the partial derivatives in (5.168), (5.170) and (5.171) 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) (5.172)
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, integrating the series in (5.163) over the interval [−π, π], which is
justified by the uniform convergence of the series, to obtain
Z π
P (r, θ) dθ = 1, for all 0 6 r < 1. (5.173)
−π
The series defining the Poisson kernel in (5.163) can actually be evaluated
by using the identity
2 cos(nθ) = einθ + e−inθ ,
and then adding geometric series. Indeed,
∞
X ∞
X
n
2r 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 − reiθ 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]. (5.174)
n=1
1 − 2r cos(θ) + r2
Substituting the value of the series in (5.174) into (5.163) then yields the formula
1 1 − r2
P (r, θ) = , for 0 6 r < 1 and θ ∈ [−π, π], (5.175)
2π 1 − 2r cos(θ) + r2
for the Poisson kernel.

We will next use the formula in (5.175) 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
(5.163) in the following proposition.
Proposition 5.2.16 (Properties of the Poisson Kernel). Let P (r, θ) be given

by (5.175), or its equivalent representation as an infinite series in (5.163). 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. (5.176)
It follows from (5.176) and the formula for P (r, θ) in (5.175) that P (r, θ) is
defined for all r ∈ [0, r) and all θ ∈ R, and P (r, θ) > 0 for r ∈ [0, r) and all
θ ∈ R; we have therefore establlished (i).
From (5.176) 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 (5.175) is not zero for
0 6 r < 1 and θ ∈ R; hence, the since the numerator and denominator of the
expression defining P (r, θ) in (5.175) are C ∞ functions, (ii) also follows.
We have already established that P satisfies Laplace’s equation in D1 (see
the calculations leading up to (5.172) on page 113) using the definition of P in
(5.163). Thus, P is harmonic in D1 and so we have established (iii).
The integral identity in (iv) will follow from (5.173) 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 (5.175) 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. (5.177)
2π 1 − r
The assertion in (vi) follows from (5.177).
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 (5.178)

r→1−
The assertion in (v) then follows from (5.178) and the expression for the Poisson
kernel in (5.175).
5.2.5 The Poisson Integral Representation

Let g : [−π, π] → R be a continuous function that can be extended to a contin-
uous 2π–periodic function in R. We then have that
|g(θ)| 6 M, for all θ ∈ [−π, π], (5.179)
and some positive constant M . The goal of this section is to use the properties
of the Poisson kernel listed in Proposition 5.2.16 to prove that the function
u : D1 → R defined by
Z π


 P (r, θ − ξ)g(ξ) dξ, for 0 6 r < 1, θ ∈ [−π, π];
−π
u(r, θ) = (5.180)



g(θ), for r = 1, θ ∈ [−π, π],
where P (r, θ) denotes the Poisson kernel for the unit disk in R2 given in (5.163)
or (5.175), solves the Dirichlet problem for the unit disk in R2 .
We first show that u ∈ C 2 (D1 ) and that it solves Laplace’s equation in D1 ;
in polar coordinates,
∂ 2 u 1 ∂u 1 ∂2u
+ + = 0, for 0 6 r < 1, θ ∈ [−π, π]. (5.181)
∂r2 r ∂r r2 ∂θ2
This will follow from (iii) in Proposition 5.2.16, provided we can show that
differentiation under the integral sign in the first part of the definition of u in
(5.180) is valid. Indeed, property (iii) in Proposition 5.181 says that
∂2P 1 ∂P 1 ∂2P
2
+ + 2 = 0, for 0 6 r < 1, θ ∈ [−π, π]. (5.182)
∂r r ∂r r ∂θ2
Thus, assuming for the moment that differentiation under the integral sign in
(5.180) is valid, we have that, for 0 6 r < 1 and θ ∈ [−π, π],
Z π 2
∂ 2 u 1 ∂u 1 ∂2u ∂
2
+ + 2 2
= 2
[P (r, θ − ξ)]g(ξ) dξ
∂r r ∂r r ∂θ −π ∂r
Z π
1 ∂
+ [P (r, θ − ξ)]g(ξ) dξ
−π r ∂r
π
1 ∂2
Z
+ [P (r, θ − ξ)]g(ξ) dξ,
−π r2 ∂θ2
Z π 2
1 ∂2

∂ 1 ∂
∆u = 2
[P (r, θ − ξ)] + [P (r, θ − ξ)] + 2 2 [P (r, θ − ξ)] g(ξ) dξ,
−π ∂r r ∂r r ∂θ
where we have used the short–hand notation, ∆u, for the Laplacian of u. The
fact that u is harmonic in D1 then follows from the previous identity and (5.182).
We will next see that differentiation under the integral sign is justified. In
order to do this, we first note that the continuity of g implies that there exists
a positive constant, M , such that
|g(θ)| 6 M, for all θ ∈ [−π, π]. (5.183)
In view of (5.183) and (5.182), in order to justify the differentiation under the
integral sign in the first part of the definition of u in (5.180), 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 (5.175) 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), (5.184)
∂θ 1 − 2r cos(θ − ξ) + r2
by virtue of the expression for the Poisson kernel in (5.175). Next, take absolute
values on both sides of (5.184) and use the estimate in (5.176) to get

∂ 2r
[P (r, θ − ξ)] 6
∂θ (1 − r)2 P (r, θ − ξ), (5.185)
where we have used the positivity of the Poisson kernel in (i) of Proposition
5.2.16. Integrating on both sides of the inequality in (5.185) form −π to π and
using property (iv) in Proposition 5.2.16 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 (5.184) 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) ∂
− [P (r, θ − ξ)],
1 − 2r cos(θ − ξ) + r2 ∂θ
so that, in view of (5.184),

∂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
(5.186)
8r2 sin2 (θ − ξ)
+ P (r, θ − ξ).
(1 − 2r cos(θ − ξ) + r2 )2
Taking absolute values on both sides of (5.186) and applying the triangle in-
equality, we obtain
2
8r2

∂ 2r

∂θ2 [P (r, θ − ξ)]6
(1 − r)2 P (r, θ − ξ) + P (r, θ − ξ), (5.187)
(1 − r)4
where we have also used the estimate in (5.176) and the positivity of the Poisson
kernel (see property (i) in Proposition 5.2.16). Integrating from −π to π on both
sides of (5.187) then yields
Z π 2
8r2

∂ 2r

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 5.2.16; 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 (5.175) 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)] = − ,
∂r 2π 1 − 2r cos(θ − ξ) + r2 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 (5.175),
∂ 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), (5.188)
∂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 (5.188), applying the triangle
inequality, and use the estimate in (5.176) to obtain

∂ 2r 2r + 1
[P (r, θ − ξ)] 6 + P (r, θ − ξ), (5.189)
∂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 5.2.16).
Integrating from −π to π on both sides of (5.189) and using property (iv) in
Proposition 5.2.16 then yields
Z π
∂
[P (r, θ − ξ)] dξ 6 2r + 2r + 1 , for 0 6 r < 0, θ ∈ R,

−π
∂r 1 − r2 (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 (5.180) is justified. We have therefore established that the function u
defined in (5.180) is in C 2 (D1 ) and satisfies Laplace’s equation. It remains to
prove that u ∈ C(D1 ) and that it satisfies the boundary conditions in problem
(5.100). This will be accomplished once we prove the following lemma:
Lemma 5.2.17 (Boundary Limits of the Poisson Integral Representation). Let

u be as given in (5.180) where g is continuous on [−π, π]. Then, for every
ζ ∈ [−pi, π],
lim |u(r, θ) − g(ζ)| = 0. (5.190)
(r,θ)→(1,ζ)
Proof: First consider the case in which ζ ∈ (−π, π).

Let ε > 0 be given. Since g is is continuous on [−π, π], there exists δ1 > 0
π
such that δ1 < , and
2
ε
|ξ − ζ| < δ1 ⇒ ξ ∈ (−π, π) and |g(ξ) − g(ζ)| < . (5.191)
3
Next, use property (iii) of the Poisson kernel in Proposition 5.2.16 to write
Z π Z π
u(r, θ) − g(ζ) = P (r, θ − ξ)g(ξ) dξ − g(ζ) P (r, θ − ξ) dξ
−π −π
Z π Z π
= P (r, θ − ξ)g(ξ) dξ − P (r, θ − ξ)g(ζ) dξ,
−π −π
so that
Z π
u(r, θ) − g(ζ) = P (r, θ − ξ)[g(ξ) − g(ζ)] dξ. (5.192)
−π
Next, take absolute values on both sides of (5.192) and use the positivity of the
Poisson kernel (see property (i) in Proposition 5.2.16) to obtain that
Z π
|u(r, θ) − g(ζ)| 6 P (r, θ − ξ)|g(ξ) − g(ζ)| dξ. (5.193)
−π
We’ll next divide the integral on the right–hand–side of (5.193) 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 (5.192) to get
Z ζ+δ1 Z ζ+δ1
ε
P (r, θ − ξ)|g(ξ) − g(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

5.2.16)
Z ζ+δ1 Z π
ε
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < P (r, θ − ξ) dξ;
ζ−δ1 3 −π
hence, by property (iv) in Proposition 5.2.16,

Z ζ+δ1
ε
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < . (5.194)
ζ−δ1 3
Next, we estimate the integral over [ζ + δ1 , π]. Using the estimate in (5.183)
and the triangle inequality we obtain
Z π Z π
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < 2M P (r, θ − ξ) dξ; (5.195)
ζ+δ1 ζ+δ1
Then, for
δ1
|θ − ζ| < , (5.196)
2
we obtain from (5.195) and the positivity of the Poisson kernel that
Z π Z π
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < 2M P (r, θ − ξ) dξ; (5.197)
ζ+δ1 θ+δ1 /2
Thus, making the change of variables ω = ξ − θ in (5.197) we get

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

Z π Z π
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < 2M P (r, ω) dω. (5.198)
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
P (r, ω) < P (r, δ1 /2), for all ω ∈ [δ1 /2, π]. (5.199)
It then follows from (5.198) and (5.199) that

Z π Z π
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < 2M P (r, δ1 /2) dω,
ζ+δ1 δ1 /2
so that
Z π
δ1
P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < 2M π− P (r, δ1 /2). (5.200)
ζ+δ1 2
Now, it follows property (v) in Proposition 5.2.16 that there exists δ2 > 0 such
that
ε
|r − 1| < δ2 ⇒ P (r, δ1 /2) < (5.201)
3M (2π − δ1 )
Thus, combining (5.200), (5.196) and (5.201) we see that
Z π
δ1 ε
|r − 1| < δ2 and |θ − ζ| < ⇒ P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < . (5.202)
2 ζ+δ1 3
Similar calculations similar to those leading to (5.202) show that there exists
δ3 > 0 such that
Z ζ−δ1
δ1 ε
|r − 1| < δ3 and |θ − ζ| < ⇒ P (r, θ − ξ)|g(ξ) − g(ζ)| dξ < . (5.203)
2 −π 3

δ1
Letting δ = min , δ2 , δ3 , we see that in view of (5.193), (5.194), (5.202)
2
and (5.203), that
|r − 1| < δ and |θ − ζ| < δ ⇒ |u(r, θ) − g(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.
Appendix A
Differentiating Under the

Integral Sign
In this appendix we preset the following result about differentiation under the
integral sign.
Proposition A.0.18 (Differentiation Under the Integral Sign). Suppose that

H : R × R × (a, b) → R is a C 1 function. Define
Z t
h(x, t) = H(x, t, s) ds, for all x ∈ R, t ∈ R.
a
∂ ∂
Assume that the functions H, [H(x, t, s)] and [H(x, t, s)] are absolutely
∂x ∂t
1
integrable over (a, b). Then, the h is C and its partial derivatives are given by
Z t
∂ ∂
[h(x, t)] = [H(x, t, s)] ds
∂x a ∂x
and Z t
∂ ∂
[h(x, t)] = H(x, t, t) + [H(x, t, s)] ds.
∂t a ∂t
Proposition A.0.18 can be viewed as a generalization of the Fundamental

Theorem of Calculus and is a special case of Leibnitz Rule.
123
124 APPENDIX A. DIFFERENTIATING UNDER THE INTEGRAL SIGN
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.
125

PDE Rumbos Math182Spring2014Notes

Uploaded by

Copyright:

Available Formats

PDE Rumbos Math182Spring2014Notes

Uploaded by

Document Information

Original Description:

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

PDE Rumbos Math182Spring2014Notes

Uploaded by

Copyright:

Available Formats

Notes on Partial Differential Equations

Preliminary Lecture Notes

2 Modeling with PDEs 7

5 Solving Linear PDEs 73

5.2.3 Expansion in Terms of Eigenfunctions . . . . . . . . . . . 103

A Differentiating Under the Integral Sign 123

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

(x, y, g(x, y)), for (x, y) ∈ ∂R,