Partial Differential Equations: T. Muthukumar Tmk@iitk - Ac.in April 18, 2014

Partial Differential Equations
T. Muthukumar
[email protected]
April 18, 2014

ii
Contents
Notations vii
1 PDE: An Introduction 1
1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Well-Posedness of PDE . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Three Basic PDE: History . . . . . . . . . . . . . . . . . . . . 11
1.4 Continuity Equation . . . . . . . . . . . . . . . . . . . . . . . 13
2 First Order PDE 17

2.1 Family Of Curves . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Linear Transport Equation . . . . . . . . . . . . . . . . . . . . 19
2.2.1 One Space Dimension . . . . . . . . . . . . . . . . . . . 19
2.2.2 Higher Dimension . . . . . . . . . . . . . . . . . . . . . 20
2.2.3 Inhomogeneous Transport Equation . . . . . . . . . . . 21
2.3 Integral Surfaces and Monge Cone . . . . . . . . . . . . . . . . 22
2.3.1 Quasi-linear Equations . . . . . . . . . . . . . . . . . . 22
2.3.2 Nonlinear Equations . . . . . . . . . . . . . . . . . . . 23
2.4 Method of Characteristics . . . . . . . . . . . . . . . . . . . . 24
2.5 Complete Integrals . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5.1 Envelopes and General Integrals . . . . . . . . . . . . . 33
2.5.2 Method Of Characteristics . . . . . . . . . . . . . . . . 34
2.6 Cauchy Problem . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.1 Quasilinear . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.2 Nonlinear . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.7 Non-smooth Solutions and Shocks . . . . . . . . . . . . . . . . 57
3 Classification by Characteristics 61
3.1 Semilinear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
iii
CONTENTS iv
3.1.1 Semilinear: Two Dimension . . . . . . . . . . . . . . . 63

3.2 Quasilinear . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.4 System of First Order PDE . . . . . . . . . . . . . . . . . . . 70
3.5 System of Second Order PDE . . . . . . . . . . . . . . . . . . 72
3.6 Invariance of Discriminant . . . . . . . . . . . . . . . . . . . . 73
3.7 Standard or Canonical Forms . . . . . . . . . . . . . . . . . . 74
3.8 Reduction to Standard Form . . . . . . . . . . . . . . . . . . . 75
4 Wave Equation 81
4.1 One Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.1.1 The Vibrating String: Derivation . . . . . . . . . . . . 81
4.1.2 Travelling Waves . . . . . . . . . . . . . . . . . . . . . 83
4.1.3 Domain of Dependence and Influence . . . . . . . . . . 88
4.1.4 Standing Waves: Separation of Variable . . . . . . . . 91
4.2 Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.1 Spherical Means . . . . . . . . . . . . . . . . . . . . . . 96
4.2.2 Odd Dimension . . . . . . . . . . . . . . . . . . . . . . 100
4.2.3 Inhomogeneous Wave equation . . . . . . . . . . . . . . 101
4.3 Eigenvalue Problem of Laplacian . . . . . . . . . . . . . . . . 103
4.3.1 In Rectangle . . . . . . . . . . . . . . . . . . . . . . . . 104
4.3.2 In Disk . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5 Heat Equation 107

5.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . 108
5.3 Heat Flow on a Bar . . . . . . . . . . . . . . . . . . . . . . . . 108
5.4 On a Circular Wire . . . . . . . . . . . . . . . . . . . . . . . . 110
5.5 Inhomogeneous Equation . . . . . . . . . . . . . . . . . . . . . 112
5.6 Steady State Equation . . . . . . . . . . . . . . . . . . . . . . 113
5.7 Fundamental Solution of Heat Equation . . . . . . . . . . . . 114
6 The Laplacian 117

6.1 Properties of Laplacian . . . . . . . . . . . . . . . . . . . . . . 118
6.2 Ill-Posedness of Cauchy Problem . . . . . . . . . . . . . . . . . 121
6.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . 122
6.4 Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . 124
6.4.1 Spherical Harmonics . . . . . . . . . . . . . . . . . . . 125
CONTENTS v
6.4.2 Properties of Harmonic Functions . . . . . . . . . . . . 130

6.5 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . 138
6.6 Perron’s Method . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.6.1 Non-existence of Solutions . . . . . . . . . . . . . . . . 141
6.6.2 Characterizing regular points . . . . . . . . . . . . . . 141
6.7 Ω with Simple Geometry . . . . . . . . . . . . . . . . . . . . . 142
6.8 Poisson Equation . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.8.1 Fundamental Solution of Laplacian . . . . . . . . . . . 151
6.8.2 Existence and Uniqueness . . . . . . . . . . . . . . . . 153
6.8.3 Green’s Function . . . . . . . . . . . . . . . . . . . . . 157
6.8.4 Green’s Function for half-space . . . . . . . . . . . . . 163
6.8.5 Green’s Function for a disk . . . . . . . . . . . . . . . . 164
6.8.6 Conformal Mapping and Green’s Function . . . . . . . 167
6.8.7 Dirichlet Principle . . . . . . . . . . . . . . . . . . . . 167
6.9 Neumann Boundary Condition . . . . . . . . . . . . . . . . . . 169
Appendices 171
A The Gamma Function 173
B Normal Vector of a Surface 175
C Implicit Function Theorem 177
D Divergence Theorem 179
E Surface Area and Volume of Disk in Rn 181
F Mollifiers and Convolution 185
G Duhamel’s Principle 187
Bibliography 189
Index 191
CONTENTS vi
Notations
Symbols
Pn ∂2
∆ i=1 ∂x2i

∂
∇ , ∂ , . . . , ∂x∂n
∂x1 ∂x2
Ω denotes an open subset of Rn , not necessarily bounded
∂Ω denotes the boundary of Ω

∂ α1 αn
Dα ∂x1 α1
. . . ∂x∂n αn and α = (α1 , . . . , αn )
Function Spaces
C k (X) is the class of all k-times (k ≥ 1) continuously differentiable functions

on X
C ∞ (X) is the class of all infinitely differentiable functions on X
Cc∞ (X) is the class of all infinitely differentiable functions on X with com-
pact support
General Conventions
Br (x) denotes the open disk with centre at x and radius r
Sr (x) denotes the circle or sphere with centre at x and radius r
wn denotes the surface area of a n-dimensional sphere of radius 1.
vii
NOTATIONS viii
Chapter 1
PDE: An Introduction
A partial differential equation (PDE) is an equation involving an unknown

function u of two or more variables and some or all of its partial derivatives.
The partial differential equation is usually a mathematical representation of
problems arising in nature, around us. The process of understanding physical
systems can be divided in to three stages:
(i) Modelling the problem or deriving the mathematical equation (in our
case it would be formulating PDE). The derivation process is usually a
result of conservation laws or balancing forces.
(ii) Solving the equation (PDE). What do we mean by a solution of the

PDE?
(iii) Studying properties of the solution. Usually, we do not end up with a

definite formula for the solution. Thus, how much information about
the solution can one extract without any knowledge of the formula?
1.1 Definitions
Recall that the ordinary differential equations (ODE) dealt with functions
of one variable, u : Ω ⊂ R → R. The subset Ω could have the interval form
(a, b). The derivative of u at x ∈ Ω is defined as
u(x + h) − u(x)
u′ (x) := lim ,
h→0 h
1
CHAPTER 1. PDE: AN INTRODUCTION 2
provided the limit exists. The derivative gives the slope of the tangent line
at x ∈ Ω. How to generalise this notion of derivative to a function u :
Ω ⊂ Rn → R? These concepts are introduced in a course on multi-variable
calculus. However, we shall jump directly to concepts necessary for us to
begin this course.
Let Ω be an open subset of Rn and let u : Ω → R be a given function.
We denote the directional derivative of u at x ∈ Ω, along a vector ξ ∈ Rn , as
∂u u(x + hξ) − u(x)
(x) = lim ,
∂ξ h→0 h
provided the limit exists. The directional derivative of u at x ∈ Ω, along
the standard basis vectors ei = (0, 0, . . . , 1, 0, . . . , 0) is called the i-th partial
derivative of u at x and is given as
∂u u(x + hei ) − u(x)
ux i = (x) = lim .
∂xi h→0 h
Similarly, one can consider higher order derivatives, as well. We now in-
troduce Schwartz’s multi-index notation for derivative, which will be used
to denote a PDE in a concise form. A multi-index α ∈ Zn+ is a n-tuple
α = (α1 , . . . , αn ) of non-negative integers and let |α| = α1 + . . . + αn . If α
and β are two multi-indices, then α ≤ β means αi ≤ βi for all 1 ≤ i ≤ n.
Also, α ± β = (α1 ± β1 , . . . , αn ± βn ), α! = α1 ! . . . αn ! and xα = xα1 1 . . . xαnn
for any x ∈ Rn . A k-degree polynomial in n variables can be represented as
X
aα x α .
|α|≤k
The partial differential operator of order α is denoted as,

∂ α1 ∂ αn ∂ |α|
Dα = . . . = .
∂x1 α1 ∂xn αn ∂x1 α1 . . . ∂xn αn
For k ∈ N, we define Dk u(x) := {Dα u(x) | |α| = k}. Thus, for k = 1, we
regard Du as being arranged in a vector,

(1,0,...,0) (0,1,0,...,0) (0,0,...,0,1)
∂ ∂ ∂
∇= D ,D ,...,D = , ,..., .
∂x1 ∂x2 ∂xn
We call this the gradient vector. The dot product of P
the gradient vector with
∂2
itself ∆ := ∇ · ∇ is called the Laplacian and ∆ := ni=1 ∂x 2 . Similarly, for
i
k = 2, we regard D2 as being arranged in a matrix form (called the Hessian

matrix),  
∂2 ∂2
2 . . . ∂x1 ∂xn
 ∂x ∂2
1
∂2 
 . . . 
D =
2

∂x 2 ∂x 1
...
∂x 2 ∂x n 
 .
 
∂2 ∂2
∂xn ∂x1
... ∂x2n n×n
The trace
P of∂ 2the Hessian matrix is called the Laplace operator, denoted as
∆ := ni=1 ∂x 2 . Note that under some prescribed order on multi-indices α,
i
k P
Dk u(x) can be regarded as a vector in Rn . Then |Dk u| := ( |α|=k |Dα u|2 )1/2 .
P P
In particular, |∇u| = ( ni=1 u2xi )1/2 and |D2 u| = ( ni,j=1 u2xi xj )1/2 .
Example 1.1. Let u(x, y) : R2 → R be u(x, y) = ax2 + by 2 . Then
∇u = (ux , uy ) = (2ax, 2by).

2 uxx uyx 2a 0
D u= =
uxy uyy 0 2b
Note that, for convenience, we can view ∇u : R2 → R2 and D2 u : R2 →
2
R4 = R2 , by assigning some ordering to the partial derivatives .
Definition 1.1.1. Let Ω be an open subset of Rn . A k-th order PDE F is a
k k−1
given map F : Rn × Rn × . . . Rn × R × Ω → R having the form

F Dk u(x), Dk−1 u(x), . . . Du(x), u(x), x = 0, (1.1.1)
for each x ∈ Ω and u : Ω → R is the unknown.

A general first order PDE is of the form F (Du(x), u(x), x) = 0 and, in
particular, for a two variable function u(x, y) the PDE will be of the form
F (ux , uy , u, x, y) = 0. If u(x, y, z) is a three variable function, then the PDE
is of the form F (ux , uy , uz , u, x, y, z) = 0. A general second order PDE is of
the form F (D2 u(x), Du(x), u(x), x) = 0.
As we know the PDE is a mathematical description of the behaviour of
the associated system. Thus, our foremost aim is to solve the PDEs for
the unknown function u, usually called the solution of the PDE. The first
expected notion of solution is as follows:
Definition 1.1.2. We say u : Ω → R is a solution (in the classical sense)
to the k-th order PDE (1.1.1),
• if u is k-times differentiable with the k-th derivative being continuous
• and u satisfies the equation (1.1.1).

Example 1.2. Consider the equation ux (x, y) = 0. If we freeze the variable y,
the equation is very much like an ODE. Integrating both sides, we would get
u(x, y) = f (y) as a solution, where f is any arbitrary function of y. Thus, the
family of solution depends on the choice of f ∈ C 1 . Similarly, the solution
of uy (x, y) = 0 is u(x, y) = g(x) for any choice of g ∈ C 1 .
Example 1.3. Consider the equation ut (x, t) = u(x, t). If we freeze the vari-
able x, the equation is very much like an ODE. Integrating both sides, we
would get u(x, t) = f (x)et as a solution, where f is any arbitrary function of
x. Thus, the family of solution depends on the choice of f .
Example 1.4. Let us solve for u in the equation uxy = 4x2 y. Unlike previous
example, the PDE here involves derivatives in both the variable. Still one
can solve this PDE for a general solution. We first integrate w.r.t x both
sides to get uy = (4/3)x3 y + f (y). Then, integrating
R again w.r.t y, we get
u(x, y) = (2/3)x3 y 2 + F (y) + g(x), where F (y) = f (y) dy.
Example 1.5. Consider the equation ux (x, y) = uy (x, y). At first look this
doesn’t look simple for solving directly. But a change of coordinates rewrites
the equation in a simpler form. If we choose the coordinates w = x + y and
z = x − y, we have ux = uw + uz and uy = uw − uz . Substituting this, we get
uz (w, z) = 0 which is the form considered in Example 1.2. Hence its solution
is u(w, z) = g(w) for g ∈ C 1 and, hence, u(x, y) = g(x + y).
Observe in the above examples that, in contrast to ODE, the family of
solutions of a PDE may be indexed by a function rather than a constant.
Exercise 1. Determine a and b so that u(x, y) = eax+by is a solution to
uxxxx + uyyyy + 2uxxyy = 0.
Exercise 2. Determine the relation between a and b if u(x, y) = f (ax + by)

is a solution to 3ux − 7uy = 0 for any differentiable function f such that
f ′ (z) 6= 0 for all real z. (Answer: a = 7b/3).
It is not necessary that the general form a solution has unique represe-
nation. Note that in the example below we have three different family of
solutions for the same PDE and we may have more!
Example 1.6. Consider the PDE ut (x, t) = uxx (x, t).
(i) Note that u(x, t) = c is a solution of the PDE, for any constant c ∈ R.
Thus, we have a family of solutions depending on c.
(ii) The function u : R2 → R defined as u(x, t) = (1/2)x2 + t + c, for a given

constant c ∈ R, is a solution of the PDE ut = uxx . Because ut = 1,
ux = x and uxx = 1. We have another family of solutions for the same
PDE.
(iii) Note that u(x, t) = eax+bt is a solution to the PDE ut = uxx if b = a2 .

Note that ut = bu, ux = au and uxx = a2 u.
As of today, there is no universal way of solving a given PDE. Thus,
the PDE’s have to be categorised based on some common properties, all of
which may be expected to have a common technique to solve. One such
classification is given below.
Definition 1.1.3. We say F is linear if (1.1.1) has the form
X
aα (x)Dα u(x) = f (x)
|α|≤k
for given functions f and aα (|α| ≤ k). If f ≡ 0, we say F is homogeneous.

F is said to be semilinear, if it is linear only in the highest order, i.e., F has
the form X
aα (x)Dα u(x) + a0 (Dk−1 u, . . . , Du, u, x) = 0.
|α|=k
We say F is quasilinear if it has the form

X
aα (Dk−1 u(x), . . . , Du(x), u(x), x)Dα u + a0 (Dk−1 u, . . . , Du, u, x) = 0,
|α|=k
i.e., the highest order derivative coefficient contains derivative only upto the
previous order. Finally, we say F is fully nonlinear if it depends nonlinearly
on the highest order derivatives.
Note that the classification is heirarchical, i.e., we have the inclusion
linear ⊂ semilinear ⊂ quasilinear ⊂ fully nonlinear.
Thus, common sense tells that we classify a PDE based on the smallest class
it sits in the heirarchy.
Example 1.7. (i) a1 (x)uxx + a2 (x)uxy + a3 (x)uyy + a4 (x)ux + a5 (x)uy =

a6 (x)u is linear.
(ii) xuy − yux = u is linear.
(iii) xux + yuy = x2 + y 2 is linear.
(iv) utt − c2 uxx = f (x, t) is linear.
(v) y 2 uxx + xuyy = 0 is linear.
(vi) ux + uy − u2 = 0 is semilinear.
(vii) ut + uux + uxxx = 0 is semilinear.
(viii) u2tt + uxxxx = 0 is semilinear.
(ix) ux + uuy − u2 = 0 is quasilinear.
(x) uux + uy = 2 is quasilinear.
(xi) ux uy − u = 0 is nonlinear.
Example 1.8 (Examples of Linear PDE). Transport Equation ut (x, t)+b·
∇x u(x, t) = 0 for some given b ∈ Rn assuming that x ∈ Rn .
Laplace Equation ∆u = 0.
Poisson Equation ∆u(x) = f (x).
Poisson Equation ∆u(x) = f (u).
Helmholtz Equation ∆u + k 2 u = 0, for a given constant k.
Heat Equation ut − ∆u = 0.
Kolmogorov’s Equation ut − A · D2 u + b · ∇u = 0, for given n × n matrix
2
A = (aij ) and b ∈ Rn . The first scalar product is in Rn and the second
is in Rn .
Wave Equation utt − ∆u = 0
General Wave Equation utt − A · D2 u + b · ∇u = 0, for given n × n matrix
2
A = (aij ) and b ∈ Rn . The first scalar product is in Rn and the second
is in Rn .
Schrödinger Equation iut + ∆u = 0.

Airy’s Equation ut + uxxx = 0.
Beam Equation ut + uxxxx = 0.
Example 1.9 (Examples of Nonlinear PDE). Inviscid Burgers’ Equation
ut + uux = 0, for x ∈ R
Eikonal Equation |∇u(x)| = f (x) is a first order nonlinear equation. This
equation arises in geometrical optics, optimal control and computer
vision etc. In fact, the name “eikon” is a greek word for image.
Hamilton-Jacobi Equation ut + H(∇u, x) = 0.
Minimal Surface Equation
!
∇u
∇· p = f (x)
1 + |∇u|2
arises in geometry. The graph of the solution u defined on the domain
Ω (say convex domain, for simplicity) has the given mean curvature
f : Ω → R. When f ≡ 0 the equation is called minimal surface
equation. This is a second order elliptic type PDE.
Image Processing A degenerate elliptic equation

∇u
u − λ∇ · = f (x)
|∇u|
is a system of three equations where the solution u : Ω ⊂ R2 → R3 ,
which measures the intensity of red, green and blue pixels iin a coloured
image. The problem is given a noisy image f : Ω → R3 , we seek a de-
noised image u by diffusing the noise in directions parallel to the image
edges. The λ > 0 is a diffusive scaling.
Monge-Ampére Equation det(D2 u) = f (x, u, ∇u) is a fully nonlinear
PDE encountered in optimal transport problems. The gradient of the
solution u, ∇u, maps optimal transportation path.
Schrödinger Equation iut + ∆u − V (u)u = 0. The solution u : Rn ×
[t0 , ∞) → C, a wavefunction, is associated with a particle of mass m
and driven by a potential V (u).
Korteweg de Vries (KdV) Equation ut + ux + uux + uxxx = 0.

Exercise 3. Classify all the important PDE listed in Example 1.8.
Exercise 4. Classify the PDEs in the hierarchy of linearity:
(i) (y − u)ux + xuy = xy + u2 .
(ii) uu2x − xuy = x2 u3 .
(iii) x2 ux + (y − x)uy = y sin u.
(iv) (sin y)ux − ex uy = ey u.
(v) ux + sin(uy ) = u.
(vi) uux + x2 uyyy + sin x = 0.

2
(vii) ux + ex uy = 0.
(viii) utt + (sin y)uyy − et cos y = 0.
(ix) x2 uxx + ex u = xuxyy .
(x) ey uxxx + ex u = − sin y + 10xuy .
(xi) y 2 uxx + ex uux = 2xuy + u.
(xii) ux uxxy + ex uuy = 5x2 ux .
(xiii) ut = k 2 (uxx + uyy ) + f (x, y, t).
(xiv) x2 uxxy + y 2 uyy − log(1 + y 2 )u = 0.
(xv) ux + u3 = 1.
(xvi) uxxyy + ex ux = y.
(xvii) uuxx + uyy − u = 0.
(xviii) uxx + ut = 3u.
Exercise 5. Rewrite the following PDE in the new coordinates v and w.
(i) ux + uy = 1 for v = x + y and w = x − y.

(ii) aut + bux = u for v = ax − bt and w = t/a where a, b 6= 0.
(iii) aux + buy = 0 for v = ax + by and w = bx − ay, where a2 + b2 > 0.
(iv) utt = c2 uxx for v = x + ct and w = x − ct.
(v) uxx + 2uxy + uyy = 0 for v = x and w = x − y.
(vi) uxx − 2uxy + 5uyy = 0 for v = x + y and w = 2x.
(vii) uxx + 4uxy + 4uyy = 0 for v = y − 2x and w = x. (should get uww = 0).
(viii) uxx + 2uxy − 3uyy = 0 for v = y − 3x and w = x + y.
1.2 Well-Posedness of PDE

We have seen through examples that a given PDE may one, many or no solu-
tion(s). The choice of our solution depends on the motivation for the study.
One needs to enforce extra side conditions to a particular solution. These
extra conditions are called boundary conditions (boundary value problem)
or whose value at initial time is known(initial value problem). It is usually
desirable to solve a well-posed problem, in the sense of Hadamard . By well-
posedness we mean that the PDE along with the boundary condition (or
initial condition)
(a) has a solution (existence)
(b) the solution is unique (uniqueness)
(c) the solution depends continuously on the data given (stability).
Any PDE not meeting the above criteria is said to be ill-posed. If too many
initial/boundary conditions are specified, then the PDE will have no solu-
tion. If too few initial/boundary conditions are specified, then the PDE will
not have unique solution. However, if the right amount of initial/boundary
conditions are specified, but at wrong places, then the solution may exist and
be unique except that that it may not depend continuously on the initial or
boundary data.
Example 1.10. The initial value problem

utt = uxx in R × (0, ∞)
u(x, 0) = ut (x, 0) = 0
has the trivial solution u(x, t) = 0. We consider the same problem with a
small change in data, i.e.,

 utt = uxx in R × (0, ∞)
u(x, 0) =0

ut (x, 0) = ε sin xε
which has the solution uε (x, t) = ε2 sin(x/ε) sin(t/ε). Since
sup{|uε (x, t) − u(x, t)|} = ε2 sup {|sin(x/ε) sin(t/ε)|} = ε2

(x,t) (x,t)
the problem is well-posed.

Example 1.11 (Ill-posed). The initial value problem

utt = −uxx in R × (0, ∞)
u(x, 0) = ut (x, 0) = 0
has the trivial solution u(x, t) = 0. We consider the same problem with a
small change in data, i.e.,

 utt = −uxx in R × (0, ∞)
u(x, 0) =0

ut (x, 0) = ε sin xε
which has the solution uε (x, t) = ε2 sin(x/ε) sinh(t/ε). Since
sup{|uεt (x, t) − ut (x, t)|} = ε sup {|sin(x/ε)|} = ε

(x,t) (x,t)
and
lim sup{|uε (x, t) − u(x, t)|} = ε lim sup {|sin(x/ε)|}

t→∞ x t→∞ x
2
= lim ε |sinh(t/ε)| = ∞,
t→∞
the problem is ill-posed. Because a small change in initial data leads to a

large change in the solution.
Exercise 6. Consider the initial value problem
ut + uxx = 0 (x, t) ∈ R × (0, ∞)
with u(x, 0) = 1.
(a) Show that u(x, t) ≡ 1 is a solution to this problem.

2
en t
(b) Show that un (x, t) = 1 + n
sin(nx) is a solution to the problem with
initial value
sin(nx)
u(x, 0) = 1 + .
n
(c) Find supx {|un (x, 0) − 1|}.
(d) Find supx {|un (x, t) − 1|}.
(e) Show that the problem is ill-posed.
The fundamental question is, given a PDE, find these extra conditions
that make a PDE well-posedness.
1.3 Three Basic PDE: History

The study of partial differential equations started as a tool to analyse the
models of physical science. The PDE’s usually arise from the physical laws
such as balancing forces (Newton’s law), momentum, conservation laws etc.
The first PDE was introduced in 1752 by d’Alembert as a model to study
vibrating strings. He introduced the one dimensional wave equation
∂ 2 u(x, t) ∂ 2 u(x, t)
= .
∂t2 ∂x2
This was then generalised to two and three dimensions by Euler (1759) and
D. Bernoulli (1762), i.e.,
∂ 2 u(x, t)
= ∆u(x, t),
∂t2
P3 ∂2
where ∆ = i=1 ∂x2i .
In physics, a field is a physical quantity associated to each point of space-

time. A field can be classified as a scalar field or a vector field according to
whether the value of the field at each point is a scalar or a vector, respec-
tively. Some examples of field are Newton’s gravitational field, Coulomb’s
electrostatic field and Maxwell’s electromagnetic field.
Given a vector field V , it may be possible to associate a scalar field u,
called potential, such that ∇u = V . Moreover, the gradient of any function
u, ∇u is a vector field. In gravitation theory, the gravity potential is the
potential energy per unit mass. Thus, if E is the potential energy of an
object with mass m, then u = E/m and the potential associated with a mass
distribution is the superposition of potentials of point masses.
The Newtonian gravitation potential can be computed to be
Z
1 ρ(y)
u(x) = dy
4π Ω |x − y|
where ρ(y) is the density of the mass distribution, occupying Ω ⊂ R3 , at y. In

1782, Laplace discovered that the Newton’s gravitational potential satisfies
the equation:
∆u = 0 on R3 \ Ω.
Thus, the operator ∆ = ∇ · ∇ is called the Laplacian and any function whose
Laplacian is zero (as above) is said to be a harmonic function.
Later, in 1813, Poisson discovered that on Ω the Newtonian potential
satisfies the equation:
−∆u = ρ on Ω.
Such equations are called the Poisson equation. The identity obtained by
Laplace was, in fact, a consequence of the conservation laws and can be gener-
alised to any scalar potential. Green (1828) and Gauss (1839) observed that
the Laplacian and Poisson equations can be applied to any scalar potential
including electric and magnetic potentials. Suppose there is a Rscalar potential
u such that V = ∇u for a vector field V and V is such that ∂γ V · ν dσ = 0
for all closed surfaces ∂γ ⊂ Γ. Then, by Gauss divergence theorem1 (cf.
Appendix D), we have
Z
∇ · V dx = 0 ∀γ ⊂ Γ.
γ
1
a mathematical formulation of conservation laws
Thus, ∇ · V = divV = 0 on Γ and hence ∆u = ∇ · (∇u) = ∇ · V = 0

on Γ. Thus, the scalar potential is a harmonic function. The study of
potentials in physics is called Potential Theory and, in mathematics, it is
called Harmonic Analysis. Note that, for any potential u, its vector field
V = ∇u is irrotational, i.e., curl(V ) = ∇ × V = 0.
Later, in 1822 J. Fourier on his work on heat flow in Théorie analytique
de la chaleur introduced the heat equation
∂u(x, t)
= ∆u(x, t),
∂t
P ∂2
where ∆ = 3i=1 ∂x 2 . The heat flow model was based on Newton’s law of
i
cooling.
Thus, by the beginning of 19th century, the three most important PDE’s
were identified.
1.4 Continuity Equation

Let us consider an ideal compressible fluid (viz. gas) occupying a bounded
region Ω ⊂ Rn (in practice, we take n = 3, but the derivation is true for
all dimensions). For mathematical precision, we assume Ω to be a bounded
open subset of Rn . Let ρ(x, t) denote the density of the fluid for x ∈ Ω at
time t ∈ I ⊂ R, for some open interval I. Mathematically, we presume that
ρ ∈ C 1 (Ω × I). We cut a region Ωt ⊂ Ω and follow Ωt , the position at time t,
as t varies in I. For mathematical precision, we will assume that Ωt have C 1
boundaries (cf. Appendix D). Now, the law of conservation of mass states
that during motion the mass is conserved and mass is the product of density
and volume. Thus, the mass of the region as a function of t is constant and
hence its derivative should vanish. Therefore,
Z
d
ρ(x, t) dx = 0.
dt Ωt
We regard the points of Ωt , say x ∈ Ωt , following the trajectory x(t) with

velocity v(x, t). We also assume that the deformation of Ωt is smooth, i.e.,
v(x, t) is continuous in a neighbourhood of Ω × I. Consider

Z Z Z !
d 1
ρ(x, t) dx = lim ρ(x, t + h) dx − ρ(x, t) dx
dt Ωt h→0 h Ωt+h Ωt
Z
ρ(x, t + h) − ρ(x, t)
= lim dx
h→0 Ω h
t
Z Z !
1
+ lim ρ(x, t + h) dx − ρ(x, t + h) dx
h→0 h Ωt+h Ωt
The first integral becomes

Z Z
ρ(x, t + h) − ρ(x, t) ∂ρ
lim dx = (x, t) dx.
h→0 Ω
t
h Ωt ∂t
The second integral reduces as,

Z Z Z

ρ(x, t + h) dx − ρ(x, t + h) dx = ρ(x, t + h) χΩt+h − χΩt
Ωt+h Ωt Ω
Z
= ρ(x, t + h) dx
Ωt+h \Ωt
Z
− ρ(x, t + h) dx.
Ωt \Ωt+h
We now evaluate the above integral in the sense of Riemann. We fix

t. Our aim is to partition the set (Ωt+h \ Ωt ) ∪ (Ωt \ Ωt+h ) with cylinders
and evaluate the integral by letting the cylinders as small as possible. To
do so, we choose 0 < s ≪ 1 and a polygon that covers ∂Ωt from outside
such that the area of each of the face of the polygon is less than s and the
faces are tangent to some point xi ∈ ∂Ωt . Let the polygon have m faces.
Then, we have x1 , x2 , . . . xm at which the faces F1 , F2 , . . . , Fm are a tangent
to ∂Ωt . Since Ωt+h is the position of Ωt after time h, any point x(t) moves to
x(t + h) = v(x, t)h. Hence, the cylinders with base Fi and height v(xi , t)h is
expected to cover our annular region depending on whether we move inward
or outward. Thus, v(xi , t)·ν(xi ) is positive or negative depending on whether
Ωt+h moves outward or inward, where ν(xi ) is the unit outward normal at
xi ∈ ∂Ωt .
Z Z X m
ρ(x, t+h) dx− ρ(x, t+h) dx = lim ρ(xi , t)v(xi , t)·ν(xi )hs.
Ωt+h \Ωt Ωt \Ωt+h s→0
i=1
Thus,
Z Z ! Z
1 h→0
ρ(x, t + h) dx − ρ(x, t + h) dx → ρ(x, t)v(x, t) · ν(x) dσ.
h Ωt+h Ωt ∂Ωt
By Green’s theorem (cf. Appendix D), we have

Z Z
d ∂ρ
ρ(x, t) dx = + div(ρv) dx.
dt Ωt Ωt ∂t
Now, using conservation of mass, we get

∂ρ
+ div(ρv) = 0 in Ω × R. (1.4.1)
∂t
Equation (1.4.1) is called the equation of continuity. In fact, any quantity
that is conserved as it moves in an open set Ω satisfies the equation of con-
tinuity (1.4.1).
Chapter 2
First Order PDE
In this chapter, we try to find the general solutions and particular of first
order PDE. A general first PDE has the form
F (∇u(x), u(x), x) = 0.
2.1 Family Of Curves

We observed in the previous chapters that solutions of PDE occur as family of
curves given by a constant or arbitrary function. In fact, by eliminating the
constant or function, via differentiation may lead to the differential equation
it solves. We now look at some family of curves which arise as a solution to
first order PDE’s.
Let A ⊂ R2 be an open subset that represents a parameter set and con-
sider
u : R2 × A → R
a two parameter family of “smooth surfaces” in R3 , u(x, y, a, b), where (a, b) ∈
A. For instance, u(x, y, a, b) = (x − a)2 + (y − b)2 is a family of circles
with centre at (a, b). Differentiate w.r.t x and y, we get ux (x, y, a, b) and
uy (x, y, a, b), respectively. Eliminating a and b from the two equations, we
get a first order PDE
F (ux , uy , u, x, y) = 0
whose solutions are the given surfaces u.
Example 2.1. Consider the family of circles
u(x, y, a, b) = (x − a)2 + (y − b)2 .
17
CHAPTER 2. FIRST ORDER PDE 18
Thus, ux = 2(x − a) and uy = 2(y − b) and eliminating a and b, we get

u2x + u2y − 4u = 0
is a first order PDE.
Example 2.2. Find the first order PDE, by eliminating the arbitrary function
f , satisfied by u.
(i) u(x, y) = xy + f (x2 + y 2 )
(ii) u(x, y) = f (x/y)
Proof. (i) Differentiating the given equation w.r.t x and y, we get
ux = y + 2xf ′ , uy = x + 2yf ′ ,
respectively. Eliminating f ′ , by multiplying y and x respectively, we
get
yux − xuy = y 2 − x2 .
(ii) Differentiating the given equation w.r.t x and y, we get

1 −x ′
ux = f ′ , , uy = f,
y y2
respectively. Eliminating f ′ , by multiplying x and y respectively, we
get
xux + yuy = 0.
Example 2.3. Find the first order PDE, by eliminating the arbitrary constants
a and b, satisfied by u
(i) u(x, y) = (x + a)(y + b)
(ii) u(x, y) = ax + by
Proof. (i) Differentiating the given equation w.r.t x and y, we get
ux = y + b, uy = x + a,
respectively. Eliminating a and b, we get
ux uy = u.
(ii) Differentiating the given equation w.r.t x and y, we get
ux = a, uy = b
respectively. Eliminating a and b, we get
xux + yuy = u.
Exercise 7. Find the first order PDE, by eliminating the arbitrary function
f , satisfied by u(x, y) = ex f (2x − y). (Answer: ux + 2uy − u = 0).
Exercise 8. Find the first order PDE, by eliminating the arbitrary function
f , satisfied by u(x, y) = e−4x f (2x − 3y). (Answer: 3ux + 2uy + 12u = 0).
2.2 Linear Transport Equation

2.2.1 One Space Dimension
Derivation
Imagine a object (say, a wave) moving on the surface (of a water) with
constant speed b. At any time instant t, every point on the wave would have
travelled a distance of bt from its initial position. At any given instant t,
let u(x, t) denote the shape of the wave in space. Let us fix a point (x0 , 0)
on the wave. Now note that the value of u(x, t) is constant along the line
x = bt + x0 or x − bt = x0 . Therefore the directional derivative of u in the
direction of (b, 1) is zero. Therefore
0 = ∇u(x, t) · (b, 1) = ut + bux .
This is a simple first order linear equation called the transport equation.
Solving
We wish to solve the transport equation ut + bux = 0 which describes the
motion of an object moving with constant speed b, as seen by a fixed observer
A.
Let us imagine that another observer B (say, on a skateboard) moving
with speed b observes the same object, moving in the direction of the object.
For B the wave would appear stationary while for A, the fixed observer,
the wave would appear to travel with speed b. What is the equation of the
motion of the object for the moving observer B? To understand this we need
to identify the coordinate system for B relative to A. Let us fix a point x
at time t = 0. After time t, the point x remains as x for the fixed observer
A, while for the moving observer B, the point x is now x − bt. Therefore
the coordinate system for B is (ξ, s) where ξ = x − bt and s = t. Let v(ξ, s)
describe the motion of the object from B’s view. Then the PDE describing
the motion of object, as seen by B, is vs (ξ, s) = 0. Therefore, v(ξ, s) = f (ξ),
for some arbitrary function f (sufficiently differentiable), is the solution from
B’s perspective. To solve the problem from A’s perspective, we observe the
relations
ut = vξ ξt + vs st = −bvξ + vs and
ux = v ξ ξ x + v s s x = v ξ .
Therefore, ut + bux = −bvξ + vs + bvξ = vs and, hence, u(x, t) = v(ξ, s) =
f (ξ) = f (x − bt).
2.2.2 Higher Dimension

We consider the homogeneous initial-value transport problem in higher di-
mensions. Given a vector b ∈ Rn , we need to find u : Rn × [0, ∞) → R
satisfying
ut (x, t) + b · ∇u(x, t) = 0 in Rn × (0, ∞) (2.2.1)
By setting a new variable y = (x, t) in Rn × (0, ∞), (2.2.1) can be rewritten
as
(b, 1) · ∇y u(y) = 0 in Rn × (0, ∞).
This means that the directional derivative of u(y) along the direction (b, 1)
is zero. Thus, u must be constant along all lines in the direction of (b, 1).
The parametric representation of a line passing through a given point (x, t) ∈
Rn × [0, ∞) and in the direction of (b, 1) is given by s 7→ (x + sb, t + s), for
all s ≥ −t. Thus, u is constant on the line (x + sb, t + s) for all s ≥ −t and,
in particular, the value of u at s = 0 and s = −t are same. Hence,
u(x, t) = u(x − tb, 0).
The procedure explained above can be formalised as below.

The equation of a line passing through (x, t) and parallel to (b, 1) is (x, t)+
s(b, 1), for all s ∈ (−t, ∞), i.e., (x + sb, t + s). Thus, for a fixed (x, t) ∈
Rn × (0, ∞), we set v(s) := u(x + sb, t + s) for all s ∈ (−t, ∞). Consequently,
dv(s) d(x + sb) ∂u d(t + s)

= ∇u(x + sb, t + s) · + (x + sb, t + s)
ds ds ∂t ds
∂u
= ∇u(x + sb, t + s) · b + (x + sb, t + s)
∂t
and from (2.2.1) we have dvds

= 0 and hence v(s) ≡ constant for all s ∈ R.
Thus, in particular, v(0) = v(−t) which implies that u(x, t) = u(x − tb, 0).
If the value of u is known at time t = 0, for instance, u(x, 0) = g(x) on
Rn × {t = 0} for a function g : Rn → R, then
u(x, t) = u(x − tb, 0) = g(x − tb).
Since (x, t) was arbitrary in Rn × (0, ∞), we have u(x, t) = g(x − tb) for
all x ∈ Rn and t ≥ 0. Thus, g(x − tb) is a classical solution to (2.2.1) if
g ∈ C 1 (Rn ). If g 6∈ C 1 (Rn ), we shall call g(x − tb) to be a weak solution of
(2.2.1).
2.2.3 Inhomogeneous Transport Equation

We now consider the inhomogeneous transport problem. Given a vector
b ∈ Rn and a function f : Rn × (0, ∞) → R, find u : Rn × [0, ∞) → R
satisfying
∂u
(x, t) + b · ∇u(x, t) = f (x, t) in Rn × (0, ∞) (2.2.2)
∂t
As before, we set v(s) := u(x + sb, t + s), for all s ∈ R, and for any given
point (x, t) ∈ Rn × (0, ∞). Thus,
dv(s) ∂u
= b · ∇u(x + sb, t + s) + (x + sb, t + s) = f (x + sb, t + s).
ds ∂t
Consider,
u(x, t) − u(x − tb, 0) = v(0) − v(−t)

Z 0
dv
= ds
−t ds
Z 0
= f (x + sb, t + s) ds
−t
Z t
= f (x + (s − t)b, s) ds.
0
Rt
Thus, u(x, t) = u(x − tb, 0) + 0
f (x + (s − t)b, s) ds solves (2.2.2).
2.3 Integral Surfaces and Monge Cone

2.3.1 Quasi-linear Equations
We begin by describing the method for first order quasi-linear PDE,
F (∇u, u, x) := b(x, u(x)) · ∇u(x) − c(x, u(x)) = 0 for x ∈ Ω,
where Ω ⊂ Rn open subset, b(x, u(x)) ∈ Rn and c(x, u(x)) ∈ R. Thus,

we have (b(x, u(x)), c(x, u(x))) · (∇u(x), −1) = 0. Finding the solution u is
equivalent to finding the surface S in Rn+1 which is graph of the solution u
of the given quasi-linear PDE, i.e.,
S = {(x, z) ∈ Ω × R | u(x) − z = 0}.
The equation of the surface S is given by G(x, z) = u(x) − z. The normal

to S, at any point, is given by the gradient of G. Hence, ∇G = (∇u(x), −1)
(cf. Appendix B). Therefore, for every point (x0 , u(x0 )) ∈ S, the coefficient
vector (b(x0 , u(x0 )), c(x0 , u(x0 ))) ∈ Rn+1 is perpendicular to the normal vec-
tor (∇u(x0 ), −1). Thus, the coefficient vector must lie on the tangent plane
at (x0 , u(x0 )) of S. Define the vector field V (x, z) = (b(x, z), c(x, z)) formed
by the coefficients of the quasi-linear PDE. Then, we note from the above
discussion that S must be such that the coefficient vector field V is tangential
to S at every point of S.
Definition 2.3.1. A curve in Rn is said to be an integral curve for a given
vector field, if the curve is tangent to the vector field at each of its point.
Similarly, a surface in Rn is said to be an integral surface for a given vector

field, if the surface is tangent to the vector field at each of its point.
In the spirit of the above definition and arguments, finding a solution
to the quasi-linear PDE is equivalent to finding an integral surface S corre-
sponding to the coefficient vector field V . We view an integral surface w.r.t
V as an union of integral curves w.r.t V .
2.3.2 Nonlinear Equations

The arguments in previous section can be carried over to a general nonlinear
first order PDE. Consider the first order nonlinear PDE, F : Rn × R × Ω → R
such that (cf. (1.1.1))
F (∇u(x), u(x), x) = 0 in Ω,
where Ω ⊂ Rn , F is given and u is the unknown to be found. We want
find a surface described by the solution u in Rn+1 . In the quasi-linear case,
using the equation, we obtained a unique direction (b, c) which is tangential,
at each point, to our desired surface S! In the non-linear case, however, we
have no such unique direction cropping out of the equation. For any point
x ∈ Ω, the point (x, z), where z = u(x) is in the solution surface. Further,
if p = ∇u(x) then we have the relation F (p, z, x) = 0. Thus, for a fixed
(x, z) ∈ Ω × R, we consider the more general equation F (p, z, x) = 0 and
denote the solution set as
V (x, z) := {p ∈ Rn | F (p, z, x) = 0}.
Therefore, solving (2.5.1) is equivalent to finding a u ∈ C 1 (Ω) such that, for
all x ∈ Ω, there is a pair (x, z) for which z = u(x) and p = ∇u(x). Every
choice of p is a possible normal vector candidate (p, −1) at (x0 , z0 ) on S.
In general, these family of normals envelope a cone with vertex at (x0 , z0 )
perpendicular to S. As p varies in V (x, z), we have a possible family of
(tangent) planes through (x0 , z0 ) given by the equation
(z − z0 ) = p · (x − x0 ),
where one of the planes is tangential to the surface S. The envelope of this
family is a cone C(x0 , z0 ), called Monge cone, with vertex at (x0 , z0 ). The
envelope of the family of planes is that surface which is tangent at each of
its point to some plane from the family.
Definition 2.3.2. A surface S in Rn+1 is said to be an integral surface if

at each point (x0 , z0 ) ∈ S ⊂ Rn × R it is tangential to the Monge cone with
vertex at (x0 , z0 ).
2.4 Method of Characteristics

We have already noted that solving a first order PDE is equivalent to finding
an integral surface corresponding to the given PDE. The integral surfaces
are usually the union of integral curves, also called the characteristic curves.
Thus, finding an integral surface boils down to finding a family of character-
istic curves. The method of characteristics gives the equation to find these
curves in the form of a system of ODE. The method of characteristics is a
technique to reduce a given first order PDE into a system of ODE and then
solve the ODE using known methods, to obtain the solution of the first order
PDE.
Let us consider the first order nonlinear PDE (2.5.1) in new independent
variables p ∈ Rn , z ∈ R and x ∈ Ω. Consequently,
F (p, z, x) = F (p1 , p2 , . . . , pn , z, x1 , x2 , . . . , xn )
is a map of 2n + 1 variable. We now introduce the derivatives (assume it
exists) of F corresponding to each variable,

∇p F = (Fp1 , . . . , Fpn )
∇x F = (Fx1 , . . . , Fxn ).
The method of characteristics reduces a given first order PDE to a system of
ODE. The present idea is a generalisation of the idea employed in the study
of linear transport equation (cf. (2.2.1)). We must choose a curve x(s) in
Ω such that we can compute u and ∇u along this curve. In fact, we would
want the curve to intersect the boundary.
We begin by differentiating F w.r.t xi in (2.5.1), we get
Xn
Fpj uxj xi + Fz uxi + Fxi = 0.
j=1
Thus, we seek to find x(s) such that

Xn
Fpj (p(s), z(s), x(s))uxj xi (x(s)) + Fz (p(s), z(s), x(s))pi (s)
j=1
+ Fxi (p(s), z(s), x(s)) = 0.
To free the above equation of second order derivatives, we differentiate pi (s)

w.r.t s,
n
dpi (s) X dxj (s)
= uxi xj (x(s))
ds j=1
ds
and set
dxj (s)
= Fpj (p(s), z(s), x(s)).
ds
Thus,
dx(s)
= ∇p F (p(s), z(s), x(s)). (2.4.1)
ds
Now substituting this in the first order equation, we get
dpi (s)
= −Fz (p(s), z(s), x(s))pi (s) − Fxi (p(s), z(s), x(s)).
ds
Thus,
dp(s)
= −Fz (p(s), z(s), x(s))p(s) − ∇x F (p(s), z(s), x(s)). (2.4.2)
ds
Similarly, we differentiate z(s) w.r.t s,
Xn
dz(s) dxj (s)
= uxj (x(s))
ds j=i
ds
n
X
= uxj (x(s))Fpj (p(s), z(s), x(s))
j=i
Thus,
dz(s)
= p(s) · ∇p F (p(s), z(s), x(s)). (2.4.3)
ds
We have 2n + 1 first order ODE called the characteristic equations of (2.5.1).
The steps derived above can be summarised in the following theorem:
Theorem 2.4.1. Let u ∈ C 2 (Ω) solve (2.5.1) and x(s) solve (2.4.1), where
p(s) = ∇u(x(s)) and z(s) = u(x(s)). Then p(s) and z(s) solve (2.4.2) and
(2.4.3), respectively, for all x(s) ∈ Ω.
We end this section with remark that for linear, semi-linear and quasi-
linear PDE one can do away with (2.4.2), the ODE corresponding to p,
because for these problems (2.4.3) and (2.4.1) form a determined system.
However, for a fully nonlinear PDE one needs to solve all the 3 ODE’s to
compute the characteristic curve. The method of characteristics may be
generalised to higher order hyperbolic PDE’s.
Remark 2.4.2. If the PDE is linear, i.e., a and b are independent of u,

then the characteristic curves are lying in the xy-plane. If the a and b are
constants (independent of both x and u) then the characteristic curves are
straight lines. In the linear case the characteristics curves will not intersect.
This is easily seen from the fact that if they intersect then, at the point of
intersection, they have the same tangent. This is not possible.
(a) In first order linear problem, the ODE reduces to, as follows: Let
F (∇u, u, x) := b(x) · ∇u(x) + c(x)u(x) = 0 x ∈ Ω.
Then, in the new variable, F (p, z, x) = b(x) · p + c(x)z. Therefore, by

(2.4.1), we have
dx(s)
= ∇p F = b(x(s)).
ds
Also,
dz(s)
= b(x(s)) · p(s) = b(x(s)) · ∇u(x(s)) = −c(x(s))z(s).
ds
(b) For a semi-linear PDE
F (∇u, u, x) := b(x) · ∇u(x) + c(x, u(x)) = 0 for x ∈ Ω,
we have
dx(s)
= ∇p F = b(x(s)).
ds
Also,
dz(s)
= b(x(s), z(s)) · p(s) = b(x(s), z(s)) · ∇u(x(s)) = −c(x(s), z(s)).
ds
(c) A quasi-linear PDE has the form

F (∇u, u, x) := b(x, u(x)) · ∇u(x) + c(x, u(x)) = 0 for x ∈ Ω.
Then, in the new variable, F (p, z, x) = b(x, z) · p + c(x, z). Therefore, by
(2.4.1), we have
dx(s)
= ∇p F = b(x(s), z(s)).
ds
Also,
dz(s)
= b(x(s), z(s)) · p(s) = b(x(s), z(s)) · ∇u(x(s)) = −c(x(s), z(s)).
ds
Example 2.4. To understand the derivation of the characteristic equation and
the geometry involved, explained in this section, let us see what is happening
two varibale first order quasi-linear equation:
a(x, y, u)ux + b(x, y, u)uy = c(x, y, u). (2.4.4)
Solving for u(x, y) in the above equation is equivalent to finding the surface
S ≡ {(x, y, u(x, y))} generated by u in R3 . If u is a solution of (2.4.4), at
each (x, y) in the domain of u,
a(x, y, u)ux + b(x, y, u)uy = c(x, y, u)
a(x, y, u)ux + b(x, y, u)uy − c(x, y, u) = 0
(a(x, y, u), b(x, y, u), c(x, y, u)) · (ux , uy , −1) = 0
(a(x, y, u), b(x, y, u), c(x, y, u)) · (∇u(x, y), −1) = 0.
But (∇u(x, y), −1) is normal to S at the point (x, y) (cf. Appendix B).
Hence, the coefficients (a(x, y, u), b(x, y, u), c(x, y, u)) are perpendicular to
the normal. Thus, the coefficients (a(x, y, u), b(x, y, u), c(x, y, u)) lie on the
tangent plane to S at (x, y, u(x, y)). Hence, finding u is equivalent to find-
ing the integral surface corresponding to the coefficient vector field V =
(a(x, y, u), b(x, y, u), c(x, y, u)).
The surface is the union of curves which satisfy the property of S. Thus,
for any curve Γ ⊂ S such that at each point of Γ, the vector V (x, y) =
(a(x, y, u), b(x, y, u), c(x, y, u) is tangent to the curve. Parametrizing the
curve Γ by the variable s, we see that we are looking for the curve Γ =
{x(s), y(s), z(s)} ⊂ R3 such that
dx dy
= a(x(s), y(s), u(x(s), y(s))), = b(x(s), y(s), u(x(s), y(s))),
ds ds
dz
and = c(x(s), y(s), u(x(s), y(s))).
ds
The three ODE’s obtained are called characteristic equations. The union
of these characteristic (integral) curves give us the integral surface.
We shall now illustrate the method of characteristics for various examples
of first order PDE.
Example 2.5 (Linear Transport Equation). We have already solved the linear
transport equation by elementary method. We solve the same now using
method of characteristics. Consider the linear transport equation in two
variable,
ut + bux = 0, x ∈ R and t ∈ (0, ∞),
where the constant b ∈ R is given. Thus, the given vector field V (x, t) =
(b, 1, 0). The characteristic equations are
dx dt dz
= b, = 1, and = 0.
ds ds ds
Solving the 3 ODE’s, we get
x(s) = bs + c1 , t(s) = s + c2 , and z(s) = c3 .
Eliminating the parameter s, we get the curves (lines) x − bt = a constant
and z = a constant. Therefore, u(x, t) = f (x − bt) is the general solution,
for an arbitrary function f .
In the inhomogeneous equation case, where the ODE corresponding to z
is not zero, we intend to seek a function Φ(x, u) such that the solution u is
defined implicitly by Φ(x, u) = a constant. Suppose there is such a function
Φ then, by setting z := u,
Φxi + Φz uxi = 0 ∀i = 1, 2, . . . , n.
Assuming Φz 6= 0, we get
−Φxi
ux i = ∀i = 1, 2, . . . , n.
Φz
If u solves b(x, u) · ∇u(x) = c(x, u) then (b(w), c(w)) · ∇w Φ = 0 is a homoge-
neous first order equation of Φ in n + 1 variables. In this case,
dw(s) dΦ(w(s))
= (b, c) and = 0.
ds ds
Example 2.6 (Inhomogeneous Transport Equation). Given a constant b ∈ R

and a function f (x, t) , we wish to solve the inhomogeneous linear transport
equation,
ut (x, t) + bux (x, t) = f (x, t), x ∈ R and t ∈ (0, ∞).
As before, the first two ODE will give the projection of characteristic curve
in the xt plane, x − bt = a constant, and the third ODE becomes
dz(s)
= f (x(s), t(s)).
ds
Let’s say we need to find the value of u at the point (x, t). The parametri-
sation of the line passing through (x, t) and (x − bt, 0) in the first variable
is x(s) = x − bt + bs and in the second variable is t(s) = s, as s varies from
0 to t. If z has to be on the integral curve, then z(s) = u(x − bt + bs, s).
Therefore, the third ODE becomes,
dz(s)
= f (x(s), s) = f (x − bt + bs, s).
ds
Integrating both sides from 0 to t, we get
Z t
f (x − b(t − s), s) ds = z(t) − z(0)
0
= u(x, t) − u(x − bt, 0).
Thus, Z t
u(x, t) = u(x − bt, 0) + f (x − b(t − s), s) ds
0
is the general solution.
Example 2.7. Let us compute the general solution (in terms of arbitrary
functions) of the first order PDE xux (x, y) + yuy (x, y) = u(x, y). The char-
acteristic equations (ODE’s) are
dx dy dz
= x(s) = y(s) and = z(s).
ds ds ds
Thus, x(s) = c1 es , y(s) = c2 es and z(s) = c3 es . Eliminating the parameter s,
we get y/x = c4 and z/x = c5 . Thus, the general solution is F (y/x, z/x) = 0
for an arbitrary function F . Explicitly,
u(x, y) = xg(y/x) or u(x, y) = yf (x/y),
for some arbitrary smooth functions f and g. Compare this answer with
Example 2.3(ii).
functions) of the first order PDE yu(x, y)ux (x, y) + xu(x, y)uy (x, y) = xy.
The characteristic equations are
dx dy dz
= yz, = xz and = xy.
ds ds ds
Hence,
dx dy
0 = x −y
ds ds
2
d(x ) d(y 2 )
= −
ds ds
d(x2 − y 2 )
= .
ds
Thus, x2 − y 2 = c1 and, similarly, x2 − z 2 = c2 . Hence, the general solution
is F (x2 − y 2 , x2 − z 2 ) = 0 for some arbitrary function F . Explicitly, for some
f and g,
u2 (x, y) = x2 + f (x2 − y 2 ) or u2 (x, y) = y 2 + g(x2 − y 2 ).
functions) of the first order PDE 2yux + uuy = 2yu2 . The characteristic
equations (ODE’s) are
dx dy dz
= 2y(s) = z(s) and = 2y(s)z 2 (s).
ds ds ds
Solving in the parametric form is quite cumbersome, because we will have a
second order nonlinear ODE of y, y ′′ (s) = 2y(y ′ )2 . However, for u 6= 0, we get
dz 2
dy
= 2y(s)z(s) solving which we get ln |z| = y 2 + c1 or z = c2 ey . Similarly,
dy y2 2
dx
= z
2y
= c22y
e
solving which we get c2 x + e−y = c3 or, equivalently,
2 2
xze−y + e−y = c3 . Thus, the general solution is
2 2
F (e−y (1 + xz), e−y z) = 0.
or 2 2
u(x, y) = f e−y (1 + xu) ey
for some arbitrary smooth functions f . Note that u ≡ 0 is a solution if we

2
choose f = 0. The characteristic curves are (1 + xu)e−y = a constant and
2
along these curves ue−y is constant.
functions) of the first order PDE ux +2xuy = u2 . The characteristic equations
(ODE’s) are
dx dy dz
=1 = 2x(s) and = z 2 (s).
ds ds ds
Solving which we get x(s) = s + c1 , y(s) = s2 + 2c1 s + c2 and z(s) =
−1/(s + c3 ). Eliminating s between x and y, we get the characteristic curves
to be y − x2 = a constant and x + 1/z = a constant. Thus, the general
solution is F (y − x2 , x + 1/z) = 0. Explicitly,
−1
u(x, y) =
x + f (y − x2 )
for some arbitrary smooth functions f .

functions) of the first order PDE yux − xuy = 2xyu. The characteristic
equations (ODE’s) are
dx dy dz
= y(s) = −x(s) and = 2xyz.
ds ds ds
To avoid cumbersome ODE, let us begin by assuming y 6= 0, then dividing
the entire equation by y, we get
dx dy dz
=1 = −x(s)/y(s) and = 2xz.
ds ds ds
Solving which we get x(s) = s + c1 , y(s) = −s2 − 2c1 s + 2c2 and |z(s)| =
2
c3 es +2c1 s . Eliminating s between x and y, we get the characteristic curves
2
to be y 2 + x2 = a constant and z = c4 ex . Thus, the general solution is
2
F (y 2 + x2 , e−x z) = 0. Explicitly,
2
u(x, y) = f (y 2 + x2 )ex

functions) of the first order PDE ux1 + ex1 ux2 + ex3 ux3 = (2x1 + ex1 )eu . The
characteristic equations (ODE’s) are
dx1 dx2 dx3 dz
=1 = e x1 = ex3 and = (2x1 + ex1 )ez(s) .
ds ds ds ds
Solving which we get x1 (s) = s + c1 , x2 (s) = ec1 es + c2 , e−x3 (s) = −s + c3 and
e−z(s) = −s2 − 2c1 s − es+c1 + c4 .
Eliminating s between x1 and x2 , we get x2 − ex1 = a constant, e−x3 + x1 = a

constant and the general solution is F (x2 − ex1 , e−x3 + x1 , e−u + x21 + ex1 ) = 0.
Explcitly,
e−u = f (x2 − ex1 , e−x3 + x1 ) − x21 − ex1
2.5 Complete Integrals

In this section, we study the form of general solutions of a first order PDE,
i.e.,
F (∇u(x), u(x), x) = 0 in Ω. (2.5.1)
Let A ⊂ Rn be an open set which is the parameter set. Let us introduce the
n × (n + 1) matrix
 
ua 1 u x 1 a 1 . . . u x n a 1
 ..  .
2
(Da u, Dxa u) :=  ... ..
.
...
. 
ua n u x 1 a n . . . u x n a n
Definition 2.5.1. A C 2 function u = u(x; a) is said to be a complete integral

in Ω × A if u(x; a) solves (2.5.1) for each a ∈ A and the rank of the matrix
2
(Da u, Dxa u) is n.
The condition on the rank of the matrix means that u(x; a) strictly de-
pends on all the n components of a.
Example 2.13. The nonlinear equation ux uy = u has a complete integral
u(x, y; a, b) = xy + ab + (a, b) · (x, y). We shall derive this solution in Exam-
ple 2.22.
Example 2.14. The nonlinear equation u2x + uy = 0 has a complete integral

u(x, y; a, b) = ax − ya2 + b.
Example 2.15. For any given f : Rn → R, the complete integral of the
Clairaut’s equation
x · ∇u + f (∇u) = u
is
u(x; a) = a · x + f (a) ∀a ∈ Rn .
Example 2.16. The complete integral of the eikonal equation |∇u| = 1 is
u(x; a, b) = a · x + b for all a ∈ S(0, 1) and b ∈ R.
Example 2.17. The Hamilton-Jacobi is a special case of the nonlinear equa-
tion where F (x, z, p) = pn + H(x, p1 , . . . , pn−1 ) where H is independent of z
and pn . For any given H : Rn → R, the complete integral of the Hamilton-
Jacobi equation
ut + H(∇u) = 0
is
u(x, t; a, b) = a · x − tH(a) + b ∀a ∈ Rn , b ∈ R.
2.5.1 Envelopes and General Integrals

Definition 2.5.2. Let Ω ⊂ Rn and A ⊂ Rm be open subsets and let u =
u(x; a) be a C 1 function of both x and a. Suppose the equation Da u(x; a) = 0
is solvable for a, as a C 1 function of x, say a = φ(x), i.e., Da u(x; φ(x)) = 0,
then v(x) := u(x; φ(x)) is the envelope of the functions {u(·; a)}a∈A .
The idea is that for each x ∈ Ω, the graph of v is tangent to the graph of
u(·; a) for a = φ(x).
Example 2.18. The complete integral of the nonlinear PDE u2 (1+ |∇u|2 ) = 1
is u(x; a) = ±(1 − |x − a|2 )1/2 with |x − a| < 1. Now, solving Da u =
x−a
± (1−|x−a|2 )1/2 = 0 for a, we get a = φ(x) := x. Thus, the envelope is v(x) =
±1.
Theorem 2.5.3. Suppose for each a ∈ A, u(·; a) is a solution to (2.5.1)

and the envelope v of u, given as v(x) = u(x, φ(x)), exists then v also solves
(2.5.1).
Proof. Since v(x) = u(x; φ(x)), for each i = 1, 2, . . . , n,

m
X
vxi (x) = uxi + uaj φjxi (x) = uxi
j=1
because Da u(x; φ(x)) = 0. Therefore,
F (∇v(x), v(x), x) = F (∇u(x; φ(x)), u(x; φ(x)), x) = 0.
Definition 2.5.4. The general integral is the envelope w of the functions

u(x; a′ , h(a′ )) where a′ := (a1 , a2 , . . . , an−1 ) and h : A′ ⊂ Rn−1 → R, provided
it exists.
Example 2.19. The envelope of the nonlinear equation ux uy = u, considered
in Example 2.13, is u(x, y; a, b) = xy + ab + (a, b) · (x, y). Let h : R → R be
defined as h(a) = a, then u(x, y; a, h(a)) = u(x, y; a, a) = xy + a2 + a(x + y).
Solve Da u = 2a + x + y = 0 which yields a = φ(x) := −(x+y) 2
. Therefore, the
envelope w(x) = u(x, y; φ(x), h(φ(x))) = −(x − y)2 /4.
Example 2.20. p Consider the eikonal equation |∇u(x, y)| = 1 in two di-
mension, i.e., u2x + u2y = 1 whose complete integral is u(x, y; a1 , a2 ) =
(x, y) · (cos a1 , sin a1 ) + a2 . Consider h ≡ 0, then u(x, y; a1 , h(a1 )) = (x, y) ·
(cos a1 , sin a1 ). Thus, solving Da1 u = −x sin a1 + y cos a1 = 0, we get
a1 = arctan(y/x). Since
1 z
cos(arctan(z)) = √ and sin(arctan(z)) = √ ,
1 + z2 1 + z2
p
we have the envelope w(x) = ± x2 + y 2 for non-zero vectors.
Example 2.21. Consider the Hamilton-Jacobi Equation ut + |∇u|2 = 0 whose
complete integral is u(x, t; a, b) = x · a − t|a|2 + b, where a ∈ Rn and b ∈ R.
Consider h ≡ 0, then u(x, t; a, h(a)) = x · a − t|a|2 . Thus, solving Da u =
2
x − 2ta = 0, we get a = 2tx
. We get the envelope w(x) = |x| 4t
.
2.5.2 Method Of Characteristics

We have already derived the characteristic equation for a first order PDE.
We illustrate through examples its application for nonlinear PDE.
Example 2.22. Let us compute a complete integral of the first order PDE
ux uy = u(x, y). The equation is of the form F (p, z, x) = p1 p2 − z. The
characteristic equations are (using (2.4.1))

dx(s) dy(s)
, = (p2 (s), p1 (s)),
ds ds
(using (2.4.2))
dp(s)
= p(s)
ds
and (using (2.4.3))
dz(s)
= (p1 (s), p2 (s)) · (p2 (s), p1 (s)) = 2p1 (s)p2 (s)
ds
Thus, on integrating, we get p1 (s) = c1 es and p2 (s) = c2 es . Solving for z, we

get
z(s) = c1 c2 e2s + c3 .
Using p, we solve for x to get x(s) = c2 es + b and y(s) = c1 es + a.Therefore,
u(x, y) = (y − a)(x − b) + c3
is a complete integral for arbitrary constants a and b, if we choose c3 = 0.

We have already seen in Example 2.13 that u(x, y) = xy + ab + (a, b) · (x, y)
is a complete integral.
Example 2.23. Let us find a different complete integral of the nonlinear PDE
ux uy = u. Note that F (p, z, x) = p1 p2 − z. Then the ODE

dx(s) dy(s)
, = (p2 (s), p1 (s)),
ds ds
dp(s)
= p(s)
ds
Thus, on integrating, we get p1 (s) = c1 es and p2 (s) = c2 es . Therefore,
√
p1 /p2 = a. Using this equation with p1 p2 = z, we get p1 = ± az and
p
p2 = ± z/a. Now,

dz(s) dx(s) dy(s)
= (p1 (s), p2 (s)) · ,
ds ds ds
√ dx(s) p dy(s)
= ± az ± z/a
ds ds
1 dz(s) √ dx(s) √ dy(s)
√ = ± a + 1/ a
z ds ds ds
√ √ √
2 z = ± ax + y/ a + c3 .
Thus,
2
1 √ √
u(x, y) = b + ax + y/ a
2
is a complete integral, if we had chosen a > 0.
Note that previous two examples compute two different complete integral
for same equation. However, in both examples, no choice of a and b will give
the zero solution u ≡ 0. Thus, u ≡ 0 is called singular solution.
Example 2.24. Let us find the complete integral, general solution and singular
solution of the fully non-linear PDE u2x + u2y = 1 + 2u. Since F (p, z, x) =
p21 + p22 − 1 − 2z, the ODEs are

dx(s) dy(s)
, = (2p1 (s), 2p2 (s)),
ds ds
dp(s)
= (2p1 , 2p2 )
ds
Thus, on dividing and integrating, we get p1 /p2 = a. Using the PDE, we get
(1 + a2 )p22 = 1 + 2z. Thus,
r r
1 + 2z 1 + 2z
p2 = ± p1 = ±a .
1 + a2 1 + a2
Now,

dz(s) dx(s) dy(s)
= (p1 (s), p2 (s)) · ,
ds ds ds
r
1 + 2z dx(s) dy(s)
= ± a +
1 + a2 ds ds

1 dz(s) 1 dx(s) dy(s)
√ = ±√ a +
1 + 2z ds 1 + a2 ds ds
√ ax + y
1 + 2z = ± √ ± b.
1 + a2
Thus, 2
1 ax + y 1
u(x, y) = √ +b −
2 1 + a2 2
is a complete integral. Note that no choice of a and b will give the constant
solution u = −1/2. Thus, u ≡ −1/2 is called singular solution.
Exercise 9. Find the general solution (in terms of arbitrary functions) of the
given first order PDE
(i) xux + yuy = xe−u with x > 0. (Answer: u(x, y) = f (y/x) for some
arbitrary f ).
(ii) ux + uy = y + u. (Answer: u(x, y) = −(1 + y) + f (y − x)ex ).
(iii) x2 ux + y 2 uy = (x + y)u. (Answer: u(x, y) = f ((1/x) − (1/y))(x − y)).
(iv) x(y 2 − u2 )ux − y(u2 + x2 )uy = (x2 + y 2 )u. (Answer: u(x, y) = xy f (x2 +
y 2 + u2 )).
(v) (ln(y + u))ux + uy = −1.
(vi) x(y − u)ux + y(u − x)uy = (x − y)u.
(vii) u(u2 + xy)(xux − yuy ) = x4 .
(viii) (y + xu)ux − (x + yu)uy = x2 − y 2 .
(ix) (y 2 + u2 )ux − xyuy + xu = 0.
(x) (y − u)ux + (u − x)uy = x − y.

(xi) x(y 2 + u)ux − y(x2 + u)uy = (x2 − y 2 )u.

√
(xii) 1 − x2 ux + uy = 0.
(xiii) (x + y)uux + (x − y)uuy = x2 + y 2 .
(xiv) Find a complete integral of uux uy = ux + uy .
(xv) Find a complete integral of u2x + u2y = xu.
2.6 Cauchy Problem

Recall that the general solution of the transport equation depends on the
value of u at time t = 0, i.e., the value of u on the curve (x, 0) in the xt-
plane. Thus, the problem of finding a function u satisfying the first order
PDE (2.4.4) such that u is known on a curve Γ in the xy-plane is called the
Cauchy problem.
Definition 2.6.1. A Cauchy problem states that: given a hypersurface Γ ⊂
Rn , can we find a solution u of F (x, u(x), ∇u(x)) = 0 whose graph contains
Γ?
The question that arises at this moment is that: Does the knowledge of u
on any hypersurface Γ ⊂ Rn lead to solving the first order PDE? The answer
is a “no”. For instance, in the transport problem, if we choose the curve
Γ = {(x, t) | x − bt = 0}, then we had no information to conclude u off the
line x − bt = 0.
2.6.1 Quasilinear
Consider the general first order qausilinear PDE with n independent variable
F (x, u, Du, D2 u) := b(x, u(x)) · Du(x) − c(u(x), x), (2.6.1)
where b = bi is a vector with n components. Let Γ ⊂ Ω be an hypersurface

defined implicitly as S(x) = 0 and ∇S 6= 0. Given u on Γ, can we calculate
all first derivatives of u on Γ? To answer this, we map Γ to a hyperplane
Γ0 by the map φ : Γ → Γ0 with its components φ = (φ1 , φ2 , . . . , φn ) such
that φn (x) = S(x) and Jφ (x) 6= 0, non-zero Jacobian of φ, for all x ∈ Ω. For
sufficiently regular hypersurfaces such a mapping exists. Note that Γ0 is a
subset of the hyperplane whose final coordinate is zero. Let y = φ(x) and
v(y) := u(x). Then
Xn
∂φk
ux i = v yk .
k=1
∂xi
Using the above equation, the first order linear PDE becomes
n
X ∂φk
bi vy = c̃(v, y), (2.6.2)
i,k=1
∂xi k
where RHS is all known on Γ0 . To understand LHS on Γ0 , note that

1
vyk (y) = lim [v(y1 , . . . , yk + h, . . . , yn−1 , 0) − v(y1 , . . . , yk , . . . , yn−1 , 0)].
h→0 h
Therefore, we know vyk , for all k = 1, 2, . . . , n − 1, on Γ0 . We only do not

know vyn on Γ0 . Thus, (2.6.2) can be rewritten as
Xn
∂φn
bi vy = terms known on Γ0 .
i=1
∂xi n
Since φn (x) = S(x), we can compute vyn if

n
X
bi (x)Sxi 6= 0
i=1
on Γ. Note that ∇S is the normal to the hypersurface S(x) = 0.

Definition 2.6.2. We say a hypersurface Γ ⊂ Ω ⊂ Rn is non-characteristic
w.r.t (2.6.1) if
Xn
bi (u(x), x)νi (x) 6= 0 ∀x ∈ Γ,
i=1
where ν(x) is the normal vector of Γ at x.
For instance, in the two dimension case, Γ = {γ1 (r), γ2 (r)} ⊂ Ω ⊂ R2 is
non-characteristic for the quasilinear Cauchy problem

a(x, y, u)ux + b(x, y, u)uy = c(x, y, u) (x, y) ∈ Ω
(2.6.3)
u =g on Γ
if Γ is nowhere tangent to (a(γ1 , γ2 , g), b(γ1 , γ2 , g)), i.e.,
(a(γ1 , γ2 , g), b(γ1 , γ2 , g)) · (−γ2′ , γ1′ ) 6= 0.
Example 2.25. Consider the equation
2ux (x, y) + 3uy (x, y) = 1 in R2 .
Let Γ be a straight line y = mx + c in R2 . The parametrisation of the line is

Γ(r) := (r, mr + c) for r ∈ R. Therefore,
(a(γ1 (r), γ2 (r)), b(γ1 (r), γ2 (r))) · (−γ2′ (r), γ1′ (r)) = (2, 3) · (−m, 1) = 3 − 2m.
Thus, the line is not a non-characteristic for m = 3/2, i.e., all lines with slope
3/2 is not a non-characteristic.
Theorem 2.6.3. Let a, b and c, the coefficients of (2.6.3), have continuous

partial derivatives w.r.t x, y, u. Let Γ(r) := (γ1 (r), γ2 (r)) be the parametriza-
tion of an initial curve on which u(γ1 (r), γ2 (r)) = φ(r) such that γ1 , γ2 and
φ are continuously differentiable and the initial curve is non-characteristic,
i.e.,
(a(γ1 , γ2 , φ), b(γ1 , γ2 , φ)) · (−γ2′ , γ1′ ) 6= 0.
Then there exists a unique solution u(x, y) in some neighbourhood of Γ which
satisfies (2.6.3).
Proof. The characteristic curves are solution to the ODE’s
dx dy du
(r, s) = a(x, y, u); (r, s) = b(x, y, u) and (r, s) = c(x, y, u)
ds ds ds
such that on Γ, x(r, s) = γ1 (r), y(r, s) = γ2 (r) and z(r, s) = φ(r). The
non-characteristic property of Γ implies that the following Jacobian on Γ:
∂x ∂x

∂r
∂y
∂s
∂y = (a(γ1 , γ2 , φ), b(γ1 , γ2 , φ)) · (−γ2′ , γ1′ ) 6= 0.
∂r ∂s
By implicit function theorem (cf. Theorem C.0.2), one can solve for r and
s in terms of x and y, locally, in the neighbourhood of Γ. Set v(x, y) =
u(r(x, y), s(x, y)). We will show that v is a unique solution to (2.6.3). Note
that, by implicit function theorem, the value of v on Γ is
v(x, y) = u(γ1 (r), γ2 (r)) = φ(r).

Moreover,
avx + bvy = a(ur rx + us sx ) + b(ur ry + us sy )

= ur (arx + bry ) + us (asx + bsy )
= ur (rx xs + ry ys ) + us (sx xs + sy ys )
dr ds du
= ur + us = =c
ds ds ds
Theorem 2.6.4. If the initial Γ is non-characteristic and there exists func-

tions α and β, as above, then there exists a unique (depending on α and β)
solution u(x, y) in some neighbourhood of Γ which satisfies (2.6.5).
Recall that we already introduced the notion of well-posedness of a PDE
in Chapter 1. We see the existence issues in a example which highlights the
importance of well-posedness of Cauchy problem. In particular, if Γ is not
non-characteristic, then the Cauchy problem is not well-posed.
Example 2.26. (i) Find the general solution (in terms of arbitrary func-
tions) of the first order PDE 2ux (x, y) + 3uy (x, y) + 8u(x, y) = 0.
(ii) For the PDE given above, check for the characteristic property of the
following curves
(a) y = x in the xy-plane

3x−1
(b) y = 2
.
(iii) Discuss the particular solutions of the above PDE, corresponding to
(a) u(x, x) = x4 on y = x
(b) u(x, (3x − 1)/2) = x2 on y = (3x − 1)/2
(c) u(x, (3x − 1)/2) = e−4x .
Observe the nature of solutions for the same PDE on a characteristic

curve and on non-characteristic curve.
(i) The characteristic equations are

dx dy dz
= 2, = 3 and = −8z.
ds ds ds
Hence,
x(s) = 2s + c1 y(s) = 3s + c2 and z(s) = c3 e−8s .
Thus, 3x − 2y = c4 and z = c4 e−4x or z = c5 e−8y/3 . Hence, the general

solution is F (3x − 2y, e4x z) = 0. Explicitly, for some f and g,
u(x, y) = f (3x − 2y)e−4x or u(x, y) = g(3x − 2y)e−8y/3 .
(ii) (a) Parametrise the curve y = x as Γ(r) : r 7→ (r, r). Thus γ1 (r) =
γ2 (r) = r. Since the coefficients of the PDE are a(r) = 2 and
b(r) = 3, we have
(a, b) · (−γ2′ (r), γ1′ (r)) = (2, 3) · (−1, 1) = −2 + 3 = 1 6= 0.
Hence Γ is non-characteristic.
(b) Parametrise the curve y = (3x − 1)/2 as Γ(r) : r 7→ (r, (3r − 1)/2).
Hence γ1 (r) = r and γ2 (r) = (3r − 1)/2 and
(a, b) · (−γ2′ (r), γ1′ (r)) = (2, 3) · (−3/2, 1) = −3 + 3 = 0.
Hence Γ is a characteristic curve.
(iii) Recall that the general solution is F (3x − 2y, e4x z) = 0 or
u(x, y) = f (3x − 2y)e−4x or u(x, y) = g(3x − 2y)e−8y/3 .
(a) Now, u(x, x) = x4 implies F (x, e4x x4 ) = 0. Thus,
e4x z = e12x−8y (3x − 2y)4
and
u(x, y) = (3x − 2y)4 e8(x−y) .
Thus, we have a unique solution u.
(b) Using the given condition, we have F (1, x2 e4x ) = 0. Either f (1) =
x2 e4x or f (x2 e4x ) = 1. The first case is not valid (multi-valued
function). The second case corresponds to z = e−4x which will not
satisfy the Cauchy data. Hence there is no solution u solving the
given PDE with the given data.
(c) Once again using the given condition, we have F (1, x2 e4x ) = 0.
Either f (1) = x2 e4x or f (x2 e4x ) = 1. The first case is not valid
(multi-valued function). The second case corresponds to z = e−4x
which will satisfy the Cauchy data. Since there many choices of f
that satisfies f (x2 e4x ) = 1, we have infinite number of solutions (or
choices for) u that solves the PDE.
tions) of the first order PDE ux (x, y) + uy (x, y) = 1.
(ii) For the PDE given above, check for the characteristic property of the
following curves
(a) the x-axis, {(x, 0)}, in the xy-plane

(b) y = x.
(iii) Discuss the particular solutions of the above PDE, corresponding to
(a) u(x, 0) = φ(x) on x-axis.

(b) u(x, x) = x on y = x.
(c) u(x, x) = 1 on y = x.

dx dy dz
= 1, = 1 and = 1.
ds ds ds
Hence,
x(s) = s + c1 y(s) = s + c2 and z(s) = s + c3 .
Thus, y − x = c4 and z − x = c5 or x − y = c4 and z − y = c5 . Hence,
for some f and g,
u(x, y) = x + f (y − x) or u(x, y) = y + g(x − y).

(ii) (a) Parametrise the curve x-axis as Γ(r) : r 7→ (r, 0). Thus γ1 (r) = r
and γ2 (r) = 0. Since the coefficients of the PDE are a(r) = 1 and
b(r) = 1, we have
(a, b) · (−γ2′ (r), γ1′ (r)) = (1, 1) · (0, 1) = 1 6= 0.
Hence Γ is non-characteristic.
(b) Parametrise the curve y = x as Γ(r) : r 7→ (r, r). Hence γ1 (r) =
r = γ2 (r) and
(a, b) · (−γ2′ (r), γ1′ (r)) = (1, 1) · (−1, 1) = −1 + 1 = 0.
Hence Γ is a characteristic curve.
(iii) Recall that the general solution is
u(x, y) = x + f (y − x) or u(x, y) = y + g(x − y).
(a) Now, u(x, 0) = φ(x) implies f (x) = φ(−x) + x or g(x) = φ(x), and
u(x, y) = y + φ(x − y).
Thus, we have a unique solution u.

(b) Using the given condition, we have f (0) = 0 or g(0) = 0. One has
many choices of function satisying these conditions. Thus, we have
infinite number of solutions (or choices for) u that solves the PDE.
(c) Once again using the given condition, we have f (0) = 1 − x or
g(0) = 1 − x for all x ∈ R. This implies f and g are not well
defined. We have no function f and g, hence there is no solution u
solving the given PDE with the given data.
For any given (smooth enough) function φ : R → R, consider the linear

transport equation

ut + bux = 0 x ∈ R and t ∈ (0, ∞)
(2.6.4)
u(x, 0) = φ(x) x ∈ R.
We know that the general solution of the transport equation is u(x, t) = u(x−
bt, 0). Thus, u(x, t) = φ(x − bt) is the unique solution of (2.6.4). We derive
the particular solution of the Cauchy problem (2.6.4) using parametrisation
of the data curve Γ. The example also shows how the information on data
curve Γ is reduced as initial condition for the characteristic ODE’s. Note that
in the example below the data curve Γ is parametrised using the variable r
and the characteristic curves is parametrised using the variable s.
Example 2.28. We shall compute the solution of the Cauchy problem (2.6.4).
We first check for non-characteristic property of Γ. Note that Γ ≡ {(x, 0)},
the x-axis of xt-plane, is the (boundary) curve on which the value of u is given.
Thus, (Γ, φ) = {(x, 0, φ(x))} is the known curve on the solution surface of u.
We parametrize the curve Γ with r-variable, i.e., Γ = {γ1 (r), γ2 (r)} =
{(r, 0)}. Γ is non-characteristic, because (b, 1) · (0, 1) = 1 6= 0. The charac-
teristic equations are:
dx(r, s) dt(r, s) dz(r, s)

= b, = 1, and =0
ds ds ds
with initial conditions,
x(r, 0) = r, t(r, 0) = 0, and z(r, 0) = φ(r).
Solving the ODE’s, we get
x(r, s) = bs + c1 (r), t(r, s) = s + c2 (r)
and z(r, s) = c3 (r) with initial conditions
x(r, 0) = c1 (r) = r
t(r, 0) = c2 (r) = 0, and z(r, 0) = c3 (r) = φ(r).

Therefore,
x(r, s) = bs + r, t(r, s) = s, and z(r, s) = φ(r).
We solve for r, s in terms of x, t and set u(x, t) = z(r(x, t), s(x, t)).
r(x, t) = x − bt and s(x, t) = t.
Therefore, u(x, t) = z(r, s) = φ(r) = φ(x − bt).

Example 2.29. Let Ω := {(x, y) ∈ R2 | x > 0, y > 0}. Let Γ := {(x, 0) | x >
0}. Consider the linear PDE

xuy (x, y) − yux (x, y) = u(x, y) in Ω
u(x, 0) = φ(x) on Γ.
The parametrisation of the initial curve is Γ(r) := (r, 0) for r > 0. Therefore,
(a(γ1 (r), γ2 (r)), b(γ1 (r), γ2 (r))) · (−γ2′ (r), γ1′ (r)) = (0, r) · (0, 1) = r 6= 0.
Hence, the given initial curve is non-characteristic. The characteristic equa-

tions are
dx(r, s) dy(r, s) dz(r, s)
= −y; = x and = z(s)
ds ds ds
with initial conditions
x(r, 0) = r, y(r, 0) = 0, and z(r, 0) = φ(r).
Note that
d2 x(r, s) d2 y(r, s)
= −x(r, s) and = −y(r, s).
ds ds
Then, x(r, s) = c1 (r) cos s + c2 (r) sin s and y(r, s) = c3 (r) cos s + c4 (r) sin s.
Using the initial condition, we get c1 (r) = r and c3 (r) = 0. Also,
dx(r, s)
0 = −y(r, 0) = |s=0 = −c1 (r) sin 0 + c2 (r) cos 0 = c2 (r).
ds
and, similarly, c4 (r) = r. Also, z(r, s) = c5 (r)es where c5 (r) = φ(r). Thus,
we have (x(r, s), y(r, s)) = (r cos s, r sin s), where r > 0 and 0 ≤ s ≤ π/2.
Hence, r = (x2 + y 2 )1/2 and s = arctan(y/x). Therefore, for any given (x, y),
we have p
u(x, y) = z(r, s) = φ(r)es = φ( x2 + y 2 )earctan(y/x) .
Example 2.30. Let Ω := {(x, y) ∈ R2 | y > 0}. Let Γ := {(x, 0) | x ∈ R}.
Consider the semi-linear PDE

ux (x, y) + uy (x, y) = u2 (x, y) in Ω
u(x, 0) = φ(x) on Γ.
The parametrisation of the initial curve is Γ(r) := (r, 0) for all r ∈ R. There-
fore,
(a(γ1 (r), γ2 (r)), b(γ1 (r), γ2 (r))) · (−γ2′ (r), γ1′ (r)) = (1, 1) · (0, 1) = 1 6= 0.

tions are
dx(r, s) dy(r, s) dz(r, s)
= 1, = 1 and = z 2 (r, s)
ds ds ds
x(r, 0) = r, y(r, 0) = 0, and z(r, 0) = φ(r).
Therefore, x(r, s) = s + c1 (r), y(r, s) = s + c2 (r) and z(r, s) = s+c−13 (r) . Using
the initial conditions, we get c1 (r) = r, c2 (r) = 0 and c3 (r) = −(φ(r))−1 .
Note that this makes sense only if φ(r) 6= 0 for all r. To overcome this
φ(r)
situation, we write z(r, s) = 1−φ(r)s . Also, we have (x(r, s), y(r, s)) = (s+r, s),
where s ≥ 0 and r ∈ R. Moreover, r = x − y and s = y. Therefore,
φ(r) φ(x−y)
u(x, y) = z(r, s) = 1−φ(r)s = 1−φ(x−y)y . Note that the non-linearity in the
z-variable, even though the equation is linear, may cause a possible blow-up
(or singularity) in the solution. For instance, even if we assume φ is bounded,
a very large value of y may induce a singularity.
Example 2.31. Consider the quasi-linear PDE

ut (x, t) + u(x, t)ux (x, t) = x in Ω
u(x, 0) = 1 on Γ.
The parametrisation of the initial curve is Γ(r) := (r, 0) for all r ∈ R. There-
fore,
(a(γ1 (r), γ2 (r)), b(γ1 (r), γ2 (r))) · (−γ2′ (r), γ1′ (r)) = (u, 1) · (0, 1) = 1 6= 0.
tions are
= z(r, s), = 1 and = x(r, s)
ds ds ds
x(r, 0) = r, t(r, 0) = 0, and z(r, 0) = 1.
Therefore, t(r, s) = s + c1 (r), x(r, s) = c2 (r)es + c3 (r)e−s and z(r, s) =
c4 (r)es + c5 (r)e−s . Using the initial conditions, we get c1 (r) = 0, c2 (r) =
(r + 1)/2 = c4 (r), c3 (r) = (r − 1)/2 and c5 (r) = (1 − r)/2. Solving for r and
s, in terms of x, t and z, we get s = t and
2x − et + e−t
r= .
et + e−t
2t 2et
Therefore, u(x, y) = x ee2t +1
−1
+ e2t+1
.
2.6.2 Nonlinear
Definition 2.6.5. We say Γ ⊂ Ω ⊂ Rn is non-characteristic for the nonlin-
ear Cauchy problem

F (∇u(x), u(x), x) = 0 x ∈ Ω
(2.6.5)
u = g on Γ
if Γ is nowhere tangent to the Monge cone, i.e., there exists function v such
that F (v(r), g(r), γ1 (r), . . . , γn (r)) = 0 and vi = gxi , for i = 1, 2, . . . , n − 1
on Γ, and satisfies
n
X
Fpi (v, g, γ1 , . . . , γn ) · ν(x) 6= 0
i=1
where ν(x) is normal to Γ at x.
In particular, in the two dimension case, Γ = {γ1 (r), γ2 (r)} ⊂ Ω ⊂ R2 is

non-characteristic for the nonlinear Cauchy problem (2.6.5) if Γ is nowhere
tangent to the Monge cone, i.e., there exists function α(r) and β(r) such that
F (α(r), β(r), g(r), γ1 (r), γ2 (r)) = 0 and g ′ (r) = α(r)γ1′ (r) + β(r)γ2′ (r) and Γ
satisfies
(Fp2 (γ1 , γ2 , g, α, β), Fp1 (γ1 , γ2 , g, α, β)) · (−γ2′ , γ1′ ) 6= 0.
Example 2.32. For a nonlinear PDE the concept P of characteristics also depend
on initial values. Consider the nonlinear PDE ni=1 u2xi = 1 with u(x′ , 0) =
φ(x′ ). Note that u = xi is a solution with φ(x′ ) = xi . However, any choice
of φ such that |∇x′ φ(x′ )|2 > 1 has no solution.
Suppose v is a solution of F (x, u, ∇u) = 0 in a neighbourhood of some
point x0 ∈ Γ.
tions) of the first order PDE xux (x, y) + 2xuuy (x, y) = u(x, y).
(ii) For the PDE given above, check if the following curves in xy-plane are
non-characteristic and discuss the particular solutions of the PDE
(a) y = 2x2 + 1 and u(x, 2x2 + 1) = 2x.

(b) y = 3x3 and u(x, 3x3 ) = 2x2 .
(c) y = x3 − 1 and u(x, x3 − 1) = x2 .


dx dy dz
= x, = 2xz and = z.
ds ds ds
Hence,
x(s) = c1 es z(s) = c2 es and y(s) = c1 c2 e2s + c3 .
Thus, y = c2 /c1 x2 + c3 and z = c2 /c1 x. Therefore, y − zx = c3 and, for

some f ,
u(x, y) = xf (y − xu).
The characteristic curves are y − xu = a constant which depends on u.
(ii) (a) Parametrise the curve y = 2x2 + 1 as Γ(r) : r 7→ (r, 2r2 + 1). Thus
γ1 (r) = r and γ2 (r) = 2r2 + 1. Since the coefficients of the PDE
are a(r) = r and b(r, u) = 4r2 , we have
(a, b) · (−γ2′ (r), γ1′ (r)) = (r, 4r2 ) · (−4r, 1) = −4r2 + 4r2 = 0.
Hence Γ is not non-characteristic. But on the characteristic curves

y − 2x2 = 1 the function u = 2x solves the PDE. Elsewhere
the solution is non-unique and there are many choices because
u(x, 2x2 + 1) = 2x implies f (1) = 2. Thus, we have infinite num-
ber of solutions (or choices for) u that solves the PDE on other
characterisitc curves.
(b) Parametrise the curve y = 3x3 as Γ(r) : r 7→ (r, 3r3 ). Hence
γ1 (r) = r and γ2 (r) = 3r3 and
(a, b)·(−γ2′ (r), γ1′ (r)) = (r, 4r3 )·(−9r2 , 1) = −9r3 +4r3 = −5r3 6= 0
for r 6= 0. Hence Γ is a non-characteristic curve. Using (u(x, 3x3 ) =

2x2 we get 2x2 = xf (3x3 − 2x3 ) which implies 2x = f (x3 ). Thus,
f (x) = 2x1/3 and u(x, y) = 2x(y − xu)1/3 or u3 (x, y) + 8x4 u = 8x3 y.
Thus, we have a unique solution. The characteristic curves are
y − 2x3 = a constant and the data u = 2x2 is given on y − 3x3 = 0.
(c) Parametrise the curve y = x3 − 1 as Γ(r) : r 7→ (r, r3 − 1). Hence

γ1 (r) = r and γ2 (r) = r3 − 1 and
(a, b) · (−γ2′ (r), γ1′ (r)) = (r, 2r3 ) · (−3r2 , 1) = −3r3 + 2r3 = −r3 6= 0
for r 6= 0. Hence Γ is a non-characteristic curve. Using (u(x, x3 −

1) = x2 we get x = f (−1). Thus f is not well defined and, hence
there is no solution. The characteristic curves are y − x3 = a con-
stant and u = x2 given on y − x3 = −1 is not a solution.
Example 2.34. We now give an example of a fully non linear PDE. Let Ω :=
{(x, y) ∈ R2 | x > 0}. Let Γ := {(0, y) | y ∈ R}. Consider the fully
non-linear PDE
ux uy = u(x, y) in Ω
u(0, y) = y 2 on Γ.
The parametrisation of the initial curve is Γ(r) := (0, r) for all r ∈ R. We
first look for the functions α and β such that
α(r)β(r) = r2 and 2r = β(r).
Solving them, we get β(r) = 2r and α(r) = r/2. Since F (p, z, x) = p1 p2 − z,

we have
Fp2 γ2′ (r) = p1 = α(r) = r/2 6= 0, for r 6= 0.
tions are (using (2.4.1))

dx(r, s) dy(r, s)
, = (p2 (r, s), p1 (r, s)),
ds ds
(using (2.4.2))
dp(r, s)
= p(r, s)
ds
and (using (2.4.3))
dz(r, s)
= (p1 (r, s), p2 (r, s)) · (p2 (r, s), p1 (r, s)) = 2p1 (r, s)p2 (r, s)
ds
r
x(r, 0) = 0, y(r, 0) = r, z(r, 0) = r2 , p1 (r, 0) = α(r) =
2
and
p2 (r, 0) = β(r) = 2r.
Thus, on integrating, we get p1 (r, s) = (r/2)es and p2 (r, s) = 2res , for all
s ∈ R. Using p, we solve for x to get x(r, s) = 2r(es − 1) and y(r, s) =
(r/2)(es + 1). Solving for z, we get
r2 2s
z(r, s) = (e + 1).
2
Solving r and s in terms of x and y, we get
4y − x x + 4y
r= and es = .
4 4y − x
(x+4y)2
Hence u(x, y) = z(r(x, y), s(x, y)) = 16
.
Example 2.35. Consider the fully non-linear PDE

ux uy = u(x, y) in R2
u(x, 1 + x) = x2 on Γ.
The parametrisation of the initial curve is Γ(r) := (r, 1 + r) for all r ∈ R.

We first look for the functions α and β such that
α(r)β(r) = r2 and 2r = α(r) + β(r).
Solving them, we get β(r) = r = α(r). Since F (p, z, x) = p1 p2 − z, we have
Fp1 γ1′ − Fp2 γ2′ (r) = p2 − p1 = β(r) − α(r) = r − r = 0, ∀r.
Hence, the given initial curve is not non-characteristic. The characteristic

equations are (using (2.4.1))

dx(r, s) dy(r, s)
, = (p2 (r, s), p1 (r, s)),
ds ds
(using (2.4.2))
dp(r, s)
= p(r, s)
ds
and (using (2.4.3))
dz(r, s)
= (p1 (r, s), p2 (r, s)) · (p2 (r, s), p1 (r, s)) = 2p1 (r, s)p2 (r, s)
ds
x(r, 0) = r, y(r, 0) = 1 + r, z(r, 0) = r2 , p1 (r, 0) = α(r) = r
and
p2 (r, 0) = β(r) = r.
Thus, on integrating, we get p1 (r, s) = res and p2 (r, s) = res , for all s ∈ R.
Using p, we solve for x to get x(r, s) = res and y(r, s) = res + 1. Solving for
z, we get z(r, s) = r2 e2s . Note that there is no unique way of solving r and s
in terms of x and y. In fact, we have three possible represenation of u, viz.,
u = x2 , u = (y − 1)2 and u = x(y − 1). This is because the Jacobian is zero.
Of these three possibilities, only u = x(y − 1) satisfies the equation.
Example 2.36. For any given λ ∈ R, consider the fully non-linear PDE
2
ux + u2y = 1 in R2
u(x, x) = λx on Γ.
The parametrisation of the initial curve is Γ(r) := (r, r) for all r ∈ R. We

first look for the functions α and β such that
α2 (r) + β 2 (r) = 1 and λ = α(r) + β(r).
We can view α(r) = cos θ and β(r) = sin θ where θ is such that cos θ +sin θ =
λ. Since F (p, z, x) = p21 + p22 − 1, we have
Fp1 γ1′ − Fp2 γ2′ (r) = 2(p1 − p2 ) = 2(cos θ − sin θ) = 0, ∀θ = π/4 + kπ,
where k = 1, 2, . . .. Hence, the given initial curve is non-characteristic for

θ 6= π/2 + kπ for all k and cos θ + sin θ = λ. The characteristic equations are
(using (2.4.1))

dx(r, s) dy(r, s)
, = (2p1 (r, s), 2p2 (r, s)),
ds ds
(using (2.4.2))
dp(r, s)
=0
ds
and (using (2.4.3))
dz(r, s)
= (p1 (r, s), p2 (r, s)) · (2p1 (r, s), 2p2 (r, s)) = 2(p21 + p22 ) = 2
ds

x(r, 0) = r, y(r, 0) = r, z(r, 0) = λr, p1 (r, 0) = α(r) = cos θ
and
p2 (r, 0) = β(r) = sin θ.
Thus, on integrating, we get p1 (r, s) = cos θ and p2 (r, s) = sin θ. Using p, we
solve for x to get x(r, s) = 2s cos θ + r and y(r, s) = 2s sin θ + r. Solving for
z, we get z(r, s) = 2s + λr. Solving r and s in terms of x and y, we get
x−y x sin θ − y cos θ
s= and r = .
2(cos θ − sin θ) sin θ − cos θ
Therefore, the general solution is
x − y + λ(y cos θ − x sin θ)
u(x, y) =
cos θ − sin θ
x − y + (cos θ + sin θ)(y cos θ − x sin θ)
=
cos θ − sin θ
x(1 − sin θ cos θ − sin2 θ) + y(cos2 θ + sin θ cos θ − 1)
=
cos θ − sin θ
= x cos θ + y sin θ.
Example 2.37. Consider the equation xux uy + yu2y = 1 with u(2r, 0) = r.
Note that F (p, z, x, y) = xp1 p2 + yp22 − 1 = 0. Thus the ODE

dx(s) dy(s)
, = (xp2 , xp1 + 2yp2 ),
ds ds
dp(s)
= −(p1 p2 , p22 )
ds
Thus, on integrating, we get p1 /p2 = a. Using the PDE, we get (xa+y)p22 = 1.
Thus,
1 a
p2 = ± √ p1 = ± √ .
xa + y xa + y
Now,

dz(s) dx(s) dy(s)
= (p1 (s), p2 (s)) · ,
ds ds ds

1 dx(s) dy(s)
= ±√ a +
xa + y ds ds
√
u(x, y) = ±2 ax + y + b.
Thus, (u(x, y) − b)2 = 4(ax + y) is a complete integral. Using the initial

data, we get (r − b)2 = 8ar. Differentiating w.r.t r, we get r = 4a + b and
eliminating r, we get b = h(a) := −2a. Hence, (u + 2a)2 = 4(ax + y). Now
solving for a in Da u = 0, we get 4(u + 2a) = 4x. Then a = φ(x) := (x − u)/2
and the solution is
2 x−u
x =4 x + 4y
2
x2 +4y
which yields u(x, y) = 2x
.
Example 2.38. Consider the equation y(u2x −u2y )+uuy = 0 with u(2y, y) = 3y.
Note that F (p, z, x, y) = y(p21 − p22 ) + zp2 = 0. Thus the ODEs become

dx(s) dy(s)
, = (2yp1 , z − 2yp2 ),
ds ds
dp(s)
= −(p1 p2 , p21 )
ds
Thus, on integrating, we get p1 −p22 = a. Using the PDE, we get ya+up2 = 0.
2
Thus, r
ay a2 y 2
p2 = − p1 = ± a + 2 .
z z
Now,

dz(s) dx(s) dy(s)
= (p1 (s), p2 (s)) · ,
ds ds ds

√ dx(s) 1 dz(s) dy(s)
∓ a = −p z + ay
ds z 2 + ay 2 ds ds
√ p
∓ ax ∓ b = − z 2 + ay 2
√
u2 (x, y) = (b + x a)2 − ay 2 .
√
Using the initial data, we get 9y 2 = (b + 2y a)2 − ay 2 which is satisfied with
b = 0 and a = 3. Thus, u(x, y)2 = 3(x2 − y 2 ).
Example 2.39. Consider the equation y(u2x − u2y ) + uuy = 0 with u(r2 , 0) =
2r. Note that F (p, z, x, y)√ = y(p21 − p22 ) + zp2 = 0. As in above example,
we get u2 (x, √
y) = (b + x a)2 − ay 2 . Using the initial conditions, we get
4r2 = (b + r2 a)2 . Differentiating this w.r.t to r and solving for r, we get

2 1 2
r =√ √ −b .
a a
√
Substituting this in the equation of r, we get b = 1/ a. This gives
2
2 1 √ 1
u = √ +x a − ay 2 = (1 + xa)2 − ay 2 .
a a
Now solving for a in Da u = 0, we get

1 √ 1 x
0 = 2 √ +x a − √ + √ − y2
a 2a a 2 a

1
y2 = x2 − 2
a
1
a = φ(x) := p .
x2 − y 2
We choose the positive root above to keep a > 0 so that all roots above made
sense. Therefore,
!2
p x y2
u2 (x, y) = x2 − y 2 1 + p −p .
x2 − y 2 x2 − y 2
Exercise 10. Find the general solution of the following PDE. Check if the
given data curve is non-characteristic or not. Also find the solution(s) (if it
exists) given the value of u on the prescribed curves.
(i) 2ut + 3ux = 0 with u(x, 0) = sin x.
(ii) ux − uy = 1 with u(x, 0) = x2 .
(iii) ux + uy = u with u(x, 0) = cos x.
(iv) ux − uy = u with u(x, −x) = sin x.

1
(v) 4ux + uy = u2 with u(x, 0) = 1+x2
.
(vi) aux + uy = u2 with u(x, 0) = cos x.
(vii) ux + 4uy = x(u + 1) with u(x, 5x) = 1.
(viii) (1 − xu)ux + y(2x2 + u)uy = 2x(1 − xu). Also, when u(0, y) = ey on

x = 0.
2
(ix) e2y ux + xuy = xu2 . Also, when u(x, 0) = ex on y = 0.
(x) ux − 2xuuy = 0. Also, when u(x, 2x) = x−1 on y = 2x and when

u(x, x3 ) = x on y = x3 .
2 2
(xi) −3ux + uy = 0 with u(x, 0) = e−x . (Answer: u(x, y) = e−(x+3y) ).
(xii) yux + xuy = x2 + y 2 with u(x, 0) = 1 + x2 and u(0, y) = 1 + y 2 . (Answer:

u(x, y) = xy + |x2 − y 2 |).
(xiii) yux + xuy = 4xy 3 with u(x, 0) = −x4 and u(0, y) = 0.
(xiv) yux + xuy = u with u(x, 0) = x3 .
(xv) ux + yuy = y 2 with u(0, y) = sin y.
(xvi) ux + yuy = u2 with u(0, y) = sin y.
(xvii) ux + yuy = u with u(x, 3ex ) = 2.
(xviii) ux + yuy = u with u(x, ex ) = ex .
(xix) ux + xuy = u with u(1, y) = φ(y).
(xx) xux + uy = 3x − u with u(x, 0) = arctan x.
(xxi) xux + uy = 0 with u(x, 0) = φ(x).
(xxii) xux + yuy = u with u(x, 1) = 2 + e−|x| .
(xxiii) xux + yuy = xe−u with u(x, x2 ) = 0.
(xxiv) xux − yuy = 0 with u(x, x) = x4 .

2
(xxv) e2y ux + xuy = xu2 with u(x, 0) = ex .
(xxvi) uux + uy = 1 with u(2r2 , 2r) = 0 for r > 0. (Answer: No solution for
y 2 > 4x).
(xxvii) (y − u)ux + (u − x)uy = x − y with u(x, 1/x) = 0.
(xxviii) x(y 2 + u)ux − y(x2 + u)uy = (x2 − y 2 )u with u(x, −x) = 1.

√
(xxix) 1 − x2 ux + uy = 0 with u(0, y) = y.
Exercise 11. Solve the equation xux + 2yuy = 0 with u(1, y) = ey . Does a
solutions exist with data on u(0, y) = g(y) or u(x, 0) = h(x)? What happens
to characteristic curves at (0, 0)?
2
Exercise 12. Solve the equation yux + xuy = 0 with u(0, y) = e−y . In which
region of the plane is the solution uniquely determined?
Exercise 13. Solve the equation ux + yuy = 0 with u(x, 0) = 1. Also, solve
the equation with u(x, 0) = x. If there is no solution, give reasons for non-
existence.
Exercise 14. (i) Find a complete integral of x(u2x + u2y ) = uux with u given
on the curve
(a) u(2y, y) = 5y
(b) u(0, r2 ) = 2r
(ii) Find a complete integral of 4uux − u3y = 0 with u given on the curve
u(0, r) = 4r.
2.7 Non-smooth Solutions and Shocks

Observe that in the Cauchy problem of Transport equation (2.6.4), suppose
we choose a discontinuous φ, for instance,
(
1 x>0
φ(x) =
0 x≤0
then u(x, t) inherits this jump continuity. In applications it is often necessary

to consider such solutions which, by our definition, is not even a differentiable,
hence, not a solution. This demand motivates us to redefine the notion of
solution to accommodate non-smooth solutions. Further, in some quasilinear
situations, eventhough we start with a smooth initial data φ, jump disconti-
nuity might occur at some positive time t. For instance, consider the Burgers’
equation given as

ut + uux = 0 x ∈ R and t ∈ (0, ∞)
u(x, 0) = φ(x) x ∈ R.
We parametrize the curve Γ with r-variable, i.e., Γ = {γ1 (r), γ2 (r)} =

{(r, 0)}. Γ is non-characteristic, because (u, 1) · (0, 1) = 1 6= 0. The charac-
teristic equations are:

= z, = 1, and =0
ds ds ds
with initial conditions,
x(r, 0) = r, t(r, 0) = 0, and z(r, 0) = φ(r).
Solving the ODE corresponding to z, we get z(r, s) = c3 (r) with initial

conditions z(r, 0) = c3 (r) = φ(r). Thus, z(r, s) = φ(r). Using this in the
ODE of x, we get
dx(r, s)
= φ(r).
ds
Solving the ODE’s, we get
x(r, s) = φ(r)s + c1 (r), t(r, s) = s + c2 (r)
x(r, 0) = c1 (r) = r and t(r, 0) = c2 (r) = 0.
Therefore,
x(r, s) = φ(r)s + r, and t(r, s) = s.
Solving r and s, in terms of x, t and z, we get s = t and r = x−zt. Therefore,
u(x, t) = φ(x − tu) is the solution in the implicit form.
Example 2.40. If the data φ(x) = c, some constant, then u(x, t) = c and the
characteristic curves are t = x/c.
Example 2.41. In the Burgers’ equation, suppose we choose the data φ(x) =
x, then
x
u(x, t) = .
1+t
Note that u has a singularity at (0, −1). This can be observed in the following
way: u takes the constant value c along the line t = x/c − 1 and all these
curves intersect at (0, −1) which means u is multiply defined at (0, −1) or,
rather undefined at (0, −1).
Example 2.42. In the Burgers’ equation, suppose we choose the data φ to be

the function 

1 x≤0
φ(x) = 1 − x 0≤x≤1


0 x ≥ 1.
Then the characteristic curves are


t + c c≤0
x = (1 − c)t 0 ≤ c ≤ 1


c c ≥ 1.
Therefore, 

1 x≤t
1−x
u(x, t) = t≤x≤1


1−t
0 x ≥ 1.
Note that for t ≤ 1 the solution behaves well, but for t ≥ 1, the characteristics
start crossing each other on the line t = x and u(x, t) takes both 0 and 1 on
the line t = x, for t ≥ 1. This situation is called the shock.
Example 2.43. In the Burgers’ equation, suppose we choose φ to be the
function 

−1 x < −1
φ(x) = x −1 ≤ x ≤ 1


1 1 < x.
Then the characteristic curves are


−t + c c < −1
x = c(t + 1) −1 ≤ c ≤ 1


t+c 1 < c.
Therefore, 

−1 x + t < −1
x
u(x, t) = −(t + 1) ≤ x ≤ (t + 1)
 t+1

1 1 < x − t.
Example 2.44. In the Burgers’ equation, suppose we choose φ to be the

function (
0 x<0
φ(x) =
1 x > 0.
Then note that (
0 x<0
u(x, t) =
1 x > t,
but there is no information of u on the wedge {0 < x < t}.
Chapter 3
Classification by
Characteristics
A general second order PDE is of the form F (D2 u(x), Du(x), u(x), x) = 0,
for each x ∈ Ω ⊂ Rn and u : Ω → R is the unknown. A Cauchy problem
poses the following: Given the knowledge of u on a smooth hypersurface
Γ ⊂ Ω can one find the solution u of the PDE? The knowledge of u on Γ is
said to be the Cauchy data.
What should be the minimum required Cauchy data for the Cauchy prob-
lem to be solved? Viewing the Cauchy problem as an initial value problem
corresponding to ODE, we know that a unique solution exists to the second
order ODE
 ′′
 y (x) + P (x)y ′ (x) + Q(x)y(x) = 0 x ∈ I
y(x0 ) = y0

y ′ (x0 ) = y0′ .
where P and Q are continuous on I (assume I closed interval of R) and for

any point x0 ∈ I. This motivates us to define the Cauchy problem for second
order PDE as:

 F (D2 u(x), Du(x), u(x), x) = 0 x∈Ω
u(x) = g(x) x ∈ Γ (3.0.1)

Du(x) · ν(x) = h(x) x ∈ Γ
where ν is the outward unit normal vector on the hypersurface Γ and g, h

are known functions on Γ.
61
CHAPTER 3. CLASSIFICATION BY CHARACTERISTICS 62
3.1 Semilinear
Consider the general second order qausilinear PDE with n independent vari-
able
F (x, u, Du, D2 u) := A(x) · D2 u − D(∇u, u, x), (3.1.1)
where A = Aij is an n × n matrix with entries Aij (x, u, ∇u), D2 u is the
2
Hessian matrix. The dot product in LHS is in Rn . Since we demand the
solution to be in C 2 , the mixed derivatives are equal and we can assume,
without loss generality that, A is symmetric. In fact if A is not symmetric,
we can replace A with As := 12 (A0 + At0 ), which is symmetric since A · D2 u =
As · D2 u.
Let Γ ⊂ Ω be an hypersurface defined implicitly as S(x) = 0 and ∇S 6= 0.
Given u and ∇u · ν on Γ, can we calculate all other derivatives of u on Γ?
To answer this, we map Γ to a hyperplane Γ0 by the map φ : Γ → Γ0 with
its components φ = (φ1 , φ2 , . . . , φn ) such that φn (x) = S(x) and Jφ (x) 6= 0,
non-zero Jacobian of φ, for all x ∈ Ω. For sufficiently regular hypersurfaces
such a mapping exists. Note that Γ0 is a subset of the hyperplane whose
final coordinate is zero. Let y = φ(x) and v(y) := u(x). Then
n
X n
X n
∂φk ∂φk ∂φl X ∂ 2 φk
ux i = v yk ux i x j = v yk yl + v yk .
k=1
∂xi k,l=1
∂xi ∂xj k=1 ∂xi ∂xj
Using the second equation the second order linear PDE becomes
n
X X n
∂φk ∂φl ∂ 2 φk
Aij vy y = D̃(∇v, v, y) − Aij vyk , (3.1.2)
i,j,k,l=1
∂xi ∂xj k l i,k=1
∂xi ∂xj
where RHS is all known on Γ0 . To understand LHS on Γ0 , note that using

the first equation, vyk (y1 , . . . , yn−1 , 0), for all k = 1, 2, . . . , n, are known in
Γ0 . Therefore, we know vyk yl , for all l = 1, 2, . . . , n − 1, on Γ0 because
1
vyk yl (y) = lim [vyk (y1 , . . . , yl + h, . . . , yn−1 , 0) − vyk (y1 , . . . , yl , . . . , yn−1 , 0)].
h→0 h
We only do not know vyn yn on Γ0 . Therefore the (3.1.2) can be rewritten as

n
X ∂φn ∂φn
Aij vy y = terms known on Γ0 .
i,j=1
∂xi ∂xj n n
Since φn (x) = S(x), we can compute vyn yn if

n
X
Aij (x)Sxi Sxj 6= 0
i,j=1
on Γ. Note that ∇S is the normal to the hypersurface S(x) = 0.
Definition 3.1.1. We say a hypersurface Γ ⊂ Ω ⊂ Rn is non-characteristic

w.r.t (3.1.1) if
Xn
Aij (x)νi (x)νj (x) 6= 0.
i,j=1
where ν(x) is the normal vector of Γ at x.
Since any real symmetric matrix can always be diagonalised, there is

a coordinate transformation T (x) such that the matrix T (x)A(x)T t (x) is
diagonal with diagonal entries, say λ1 , λ2 , . . . , λn , for each x ∈ Ω. Since A(x)
is real symmetric all λi ∈ R, for all i. Thus, we classify PDE at a point
x ∈ Ω based on the eigenvalues of the matrix A(x). Let p denote the number
of eigenvalues that are strictly positive and z denote the number of zero
eigenvalues.
Definition 3.1.2. We say a PDE is hyperbolic at a point x ∈ Ω, if z = 0

and either p = 1 or p = n − 1. We say it is parabolic if z > 0. We say it is
elliptic, if z = 0 and either p = n or p = 0. If z = 0 and 1 < p < n − 1 then
the PDE is said to be ultra hyperbolic.
Note that the elliptic case corresponds to the situation that every hy-
persurface S(x) = 0 with ∇S 6= 0 is non-characteristic corresponding the
elliptic operator, i.e., there are no real characteristics curves. Thus, one can
equivalently define a linear second order PDE to be elliptic at x if
n
X
Aij (x)ξi ξj 6= 0 ∀ξ ∈ Rn \ {0}.
i,j=1
3.1.1 Semilinear: Two Dimension

Consider the Cauchy problem (3.0.1) in two variables and set x = (x, y). Let
τ denote the unit tangent vector on Γ. Then, the directional derivative along
the tangent vector, Du(x, y) · τ (x, y) = g ′ (x, y) is known because g is known.

Thus, we may compute directional derivative of u in any direction, along Γ,
as a linear combination of Du · τ and Du · ν. Using this we may reformulate
(3.0.1) as 
 F (D2 u, Du, u, x, y) = 0
 (x, y) ∈ Ω

u(x, y) = g(x, y) (x, y) ∈ Γ

 ux (x, y) = h1 (x, y) (x, y) ∈ Γ

uy (x, y) = h2 (x, y) (x, y) ∈ Γ
with the compatibility condition that ġ(s) = h1 γ̇1 (s) + h2 γ̇2 (s), where s 7→
(γ1 (s), γ2 (s)) is the parametrisation1 of the hypersurface Γ. The compatibil-
ity condition is an outcome of the fact that
u̇(s) = ux γ̇1 (s) + uy γ̇2 (s).
The above condition implies that among g, h1 , h2 only two can be assigned
independently.
Consider the Cauchy problem for the second order semi-linear PDE in
two variables (x, y) ∈ Ω ⊂ R2 ,


 A(x, y)uxx + 2B(x, y)uxy + C(x, y)uyy = D (x, y) ∈ Ω

u(x, y) = g(x, y) (x, y) ∈ Γ
(3.1.3)

 u x (x, y) = h1 (x, y) (x, y) ∈ Γ

uy (x, y) = h2 (x, y) (x, y) ∈ Γ.
where D(x, y, u, ux , uy ) may be non-linear and Γ is a smooth2 curve in Ω.
Also, one of the coefficients A, B or C is identically non-zero (else the PDE
is not of second order). Let s 7→ (γ1 (s), γ2 (s)) be a parametrisation of the
curve Γ. Then we have the compatibility condition that
ġ(s) = h1 γ̇1 (s) + h2 γ̇2 (s).
By computing the second derivatives of u on Γ and considering uxx , uyy
and uxy as unknowns, we have the linear system of three equations in three
unknowns on Γ,
Auxx +2Buxy +Cuyy =D
γ̇1 (s)uxx +γ̇2 (s)uxy = ḣ1 (s)
γ̇1 (s)uxy +γ̇2 (s)uyy = ḣ2 (s).
1
(′) denotes the derivative with respect to space variable and (·) denotes the derivative
with respect to parameter
2
twice differentiable
This system of equation is solvable if the determinant of the coefficients are

non-zero, i.e.,
A 2B C

γ̇1 γ̇2 0 =
6 0.
0 γ̇1 γ̇2
Definition 3.1.3. We say a curve Γ ⊂ Ω ⊂ R2 is characteristic (w.r.t

(3.1.3)) if
Aγ̇22 − 2B γ̇1 γ̇2 + C γ̇12 = 0.
where (γ1 (s), γ2 (s)) is a parametrisation of Γ.
Note that the geometry hidden in the above definition is very similar
to that we encountered in first order equation. Since ν = (−γ̇2 , γ̇1 ) is the
normal to Γ at each point, the above definition says that the curve is non-
characteristic if
2
X
Aij νi νj = Aγ̇22 − 2B γ̇1 γ̇2 + C γ̇12 6= 0
i,j=1
where A11 = A, A12 = A21 = B and A22 = C. If y = y(x) is a representation

of the curve Γ (locally, if necessary), we have γ1 (s) = s and γ2 (s) = y(s).
Then the characteristic equation reduces as
2
dy dy
A − 2B + C = 0.
dx dx
Therefore, the characteristic curves of (3.1.3) are given by the graphs whose
equation is √
dy B ± B 2 − AC
= .
dx A
Thus, we have three situations arising depending on the sign of the dis-
criminant, B 2 − AC. This classifies the given second order PDE based on
the sign of its discriminant d = B 2 − AC.
Definition 3.1.4. We say a second order PDE is of
(a) hyperbolic type if d > 0,
(b) parabolic type if d = 0 and

(c) elliptic type if d < 0.
The hyperbolic PDE have two families of characteristics, parabolic PDE

has one family of characteristic and elliptic PDE have no characteristic. We
caution here that these names are no indication of the shape of the graph
of the solution of the PDE. The classification tells us the right amount of
initial/boundary condition to be imposed for a PDE to be well-posed. For
hyperbolic, which has two real characteristics, requires as many initial con-
dition as the number of characteristics emanating from initial time and as
many boundary conditions as the number of characteristics that pass into the
spatial boundary. For parabolic, which has exactly one real characteristic,
we need one boundary condition at each point of the spatial boundary and
one initial condition at initial time. For elliptic, which has no real charac-
teristic curves, we need one boundary condition at each point of the spatial
boundary.
Note that the classification depends on the determinant of the coefficient
matrix
A B
B C
For every (x, y) ∈ Ω, the matrix is symmetric and hence diagonalisable.
If λ1 , λ2 are the diagonal entries, then d = −λ1 λ2 . Thus, a equation is
hyperbolic at a point (x, y) if the eigen values have opposite sign. It is ellipic
if the eigenvalues have same sign and is parabolic if, at least, one of the
eigenvalue is zero.
3.2 Quasilinear
All the arguments of semilinear PDE can be carried over to a quasilinear
PDE A(x, u(x), Du(x)). For each specific point x0 ∈ Ω, u(x0 ) = u0 and A0 =
A(x0 , u0 , ∇u(x0 )). The solutions to characteristic equation for a quasilinear
equation depends on the solution considered. Set
U := {(x, z, p) | x ∈ Ω, z ∈ R, p ∈ Rn }.
Definition 3.2.1. A quasilinear equation
A(x, u(x), Du(x)) · D2 u(x) = D(x, u(x), Du(x))

is said to be elliptic if the matrix Aij (x, z, p) is positive definite for each
(x, z, p) ∈ U . Further,
n
X
2
0 < α(x, z, p)|ξ| ≤ Aij (x, z, p)ξi ξj ≤ β(x, z, p)|ξ|2 ∀ξ ∈ Rn .
i,j=1
The bounds α(x, z, p) and β(x, z, p) are minimum and maximum eigen-
values, respectively. If β/α is uniformly bounded in U then PDE is uniformly
elliptic. The interesting thing about uniformly elliptic equation is that they
behave very similar to linear elliptic equations.
Example 3.1. Consider the minimal surface equation
!
∇u
∇· p = f (x)
1 + |∇u|2
where the second order coefficients are

2 −1/2 pi pj
Aij (x, z, p) = (1 + |p| ) δij −
1 + |p|2
and
1 1
α(x, z, p) = 2 3/2
β(x, z, p) = .
(1 + |p| ) (1 + |p|2 )1/2
Thus, the equation is not uniformly elliptic.
The minimal surface equation and the capillary equation are not uni-
formly elliptic.
3.3 Examples
Example 3.2 (Wave Equation). For a given c ∈ R, uyy − c2 uxx = 0 is hyper-
bolic. Since A = −c2 , B = 0 and C = 1, we have d = B 2 −AC = c2 > 0. The
eigen values of the coefficient matrix are 1, −c2 which have opposite sign.
Example 3.3 (Heat Equation). For a given c ∈ R, uy − cuxx = 0 is parabolic.
Since A = −c, B = 0 and C = 0, thus d = B 2 − AC = 0. The eigen values
of the coefficient matrix are 0, −c has a zero eigenvalue.
Example 3.4 (Laplace equation). uxx +uyy = 0 is elliptic. Since A = 1, B = 0
and C = 1, thus d = B 2 − AC = −1 < 0. The eigen values of the coefficient
matrix are 1, 1 which have same sign.
Example 3.5 (Velocity Potential Equation). In the equation (1 − M 2 )uxx +

uyy = 0, A = (1 − M 2 ), B = 0 and C = 1. Then d = B 2 − AC = −(1 − M 2 ).
The eigen values of the coefficient matrix are 1 − M 2 , 1. Thus, for M > 1
(opposite sign), the equation is hyperbolic (supersonic flow), for M = 1 (zero
eigenvalue) it is parabolic (sonic flow) and for M < 1 (same sign) it is elliptic
(subsonic flow).
Note that the classification of PDE is dependent on its coefficients. Thus,
for constant coefficients the type of PDE remains unchanged throughout
the region Ω. However, for variable coefficients, the PDE may change its
classification from region to region.
Example 3.6. An example is the Tricomi equation , uxx + xuyy = 0. The
discriminant of the Tricomi equation is d = −x. The eigenvalues are 1, x.
Thus, tricomi equation is hyperbolic when x < 0, elliptic when x > 0 and
degenerately parabolic when x = 0, i.e., y-axis. Such equations are called
mixed type.
The notion of classification of second order semi-linear PDE, discussed in
this section, could be generalised to quasi-linear, non-linear PDE and system
of ODE. However, in these cases the classification may also depend on the
solution u, as seen in the examples below.
Example 3.7. Consider the quasi-linear PDE uxx − uuyy = 0. The discrimi-
nant is d = u. The eigenvalues are 1, −u(x). It is hyperbolic for {u > 0}3 ,
elliptic when {u < 0} and parabolic when {u = 0}.
Example 3.8. Consider the quasi-linear PDE
(c2 − u2x )uxx − 2ux uy uxy + (c2 − u2y )uyy = 0
where c > 0. Then d = B 2 − AC = c2 (u2x + u2y − c2 ) = c2 (|∇u|2 − c2 ). It is

hyperbolic if |∇u| > c, parabolic if |∇u| = c and elliptic if |∇u| < c.
Example 3.9. Find the family of characteristic curves for the following second
order PDE, whenever they exist.
(i) For a given c ∈ R, uyy − c2 uxx = 0.
(ii) For a given c ∈ R, uy − cuxx = 0.
(iii) uxx + uyy = 0.

3
The notation {u > 0} means {x ∈ Ω | u(x) > 0}
(iv) uxx + xuyy = 0.
Solution. (i) We have already seen the equation is hyperbolic and hence
it should have two characteristic curves. The characteristic curves are
given by the equation
√ √
dy B ± B 2 − AC ± c2 ∓1
= = 2
= .
dx A −c c
Thus, cy ± x = a constant is the equation of the two characteristic
curves. Note that the characteristic curves y = ∓x/c + y0 are boundary
of two cones in R2 with vertex at (0, y0 ).
(ii) We have already seen the equation is parabolic and hence it should
have one characteristic curve. The characteristic curve are given by the
equation √
dy B ± B 2 − AC
= = 0.
dx A
Thus, y = a constant is the equation of the characteristic curve. i.e.,
any horizontal line in R2 is a charateristic curve.
(iii) We have already seen the equation is elliptic and hence has no real
characteristics.
(iv) The equation is of mixed type. In the region x > 0, the characteristic
curves are y ∓ 2x3/2 /3 = a constant.
Exercise 15. Classify the following second order PDE in terms of the number
of characteristics:
(a) 3uxx + uxy + 2uyy = 0.
(b) uzz + uz + urr + 1r uθ + c = 0.
(c) ut + βux + αuxx = 0.
(d) 4uxx + y 2 ux + xux + uyy + 4uxy − 4xy = 0.
(e) xuxx + xuxy + yuyy = 0.
(f) xuxx + yuxy + c = 0.

(g) x2 yuxx + xyuxy − y 2 uyy = 0.
(h) sin xuxx + 2 cos xuxy + sin xuyy = 0.
(i) uxx + 4uxy + 5uyy + ux + 2uy = 0.
(j) uxx − 4uxy + 4uyy + 3ux + 4u = 0.
(k) uxx + 2uxy − 3uyy + 2ux + 6uy = 0.
(l) (1 + x)uxx + 2xyuxy − y 2 uyy = 0.
(m) 2uxx − 4uxy + 7uyy − u = 0.
(n) uxx − 2 cos xuxy − sin2 xuyy = 0.
(o) yuxx + 2(x − 1)uxy − (y + 2)uyy = 0.
(p) yuxx + uxy − x2 uyy − ux − u = 0.
Exercise 16. Classify the following second order PDE, in terms of the number
of characteristics, and find their characteristics, when it exists:
(a) uxx + (5 + 2y 2 )uxy + (1 + y 2 )(4 + y 2 )uyy = 0.
(b) yuxx + uyy = 0.
(c) yuxx = xuyy .
(d) uyy − xuxy + yux + xuy = 0.
(e) y 2 uxx + 2xyuxy + x2 uyy = 0.
(f) uxx + 2xuxy + (1 − y 2 )uyy = 0.
3.4 System of First Order PDE

Recall that any second order ODE y ′′ = f (x, y, y ′ ) can be equivalently written
as a system of ODE, by using u1 = y and u2 = y ′ , as follows:
′
u1 = u2
′
u2 = f (x, u1 , u2 ).
A similar procedure also makes a second order PDE into a system of first
order PDEs. Thus, we expect that our classification of second order PDE to
induce a classification for a system of first order PDE.
A general system of m first order linear PDE in n variables will be of the
form
X n
Aj (x, u)uxj = f (x, u) in Ω, (3.4.1)
j=1
where each Aj (x, u) is a m × m matrix, u(x) = (u1 , . . . , um ) and f =

(f1 , . . . , fm ) has m components. Following the arguments for semilinear PDE
and using the map φ(x) and v(y) between Γ and Γ0 , we get
n
X
Aj (x, u)vyn Sxj = terms known on Γ0 .
j=1
This system is solvable for vyn if

n
!
X
det Aj (x, u)Sxj 6= 0
j=1
on Γ.
Definition 3.4.1. We say a hyperspace Γ in Rn is non-characteristic if

n
!
X
det Aj (x, u)νj (x) 6= 0
j=1
on Γ, where ν(x) = (ν1 , . . . , νn ) is the normal at x of Γ.
Definition 3.4.2. The system of first order PDE (3.4.1) is called hyperbolic
if the m × m matrix
Xn
ξj Aj (x, u)
j=1
is diagonalisable for every ξ = (ξj ) ∈ Rn and every x ∈ Ω. If the eigen-

values are all distinct, for non-zero ξ ∈ Rn , then the PDE is called strictly
hyperbolic.
Definition 3.4.3. The system of first order PDE (3.4.1) is called elliptic if
n
!
X
det Aj (x, u)ξj = 0
j=1
only for ξ = 0.
Example 3.10 (Beltrami Equations). Consider the system of first order equa-
tions

W (x, y)ux (x, y) − b(x, y)vx (x, y) − c(x, y)vy (x, y) = 0
W (x, y)uy (x, y) + a(x, y)vx (x, y) + b(x, y)vy (x, y) = 0
6 0 and the matrix
where W, a, b, c are given such that W =

a b
b c
is positive definite. Set

W −b 0 −c
A1 = A2 = .
0 a W b
Then the system can be rewritten as

ux uy 0
A1 + A2 = .
vx vy 0
Then
n
!
X W ξ1 −bξ1 − cξ2
det Aj (x, u)ξj = = W (aξ12 + 2bξ1 ξ2 + cξ22 ) 6= 0

W ξ2 aξ1 + bξ2
j=1
if ξ 6= 0. Therefore, the Beltrami equation is elliptic. The Beltrami system

is a generalization of the Cauchy-Riemann equations.
3.5 System of Second Order PDE

A general system of m second order linear PDE in n variables will be of the
form
Xn
Aij (x, u, ∇u)uxi xj + lower order terms = 0 in Ω, (3.5.1)
i,j=1
where each Aij is a m × m matrix and u(x) = (u1 , . . . , um ). Following the

arguments for semilinear PDE and using the map φ(x) and v(y) between Γ
and Γ0 , we get
n
X
Aij (x, u, ∇u)vyn yn Sxi Sxj = terms known on Γ0 .
i,j=1
Definition 3.5.1. The system is called elliptic if

n
!
X
det Aij ξi ξj = 0
i,j=1
only for ξ = 0.
3.6 Invariance of Discriminant

The classification of second order semi-linear PDE is based on the discrimi-
nant B 2 − AC. In this section, we note that the classification is independent
of the choice of coordinate system (to represent a PDE). Consider the two-
variable semilinear PDE
A(x, y)uxx +2B(x, y)uxy +C(x, y)uyy = D(x, y, u, ux , uy ) (x, y) ∈ Ω (3.6.1)
where the variables (x, y, u, ux , uy ) may appear non-linearly in D and Ω ⊂ R2 .

Also, one of the coefficients A, B or C is identically non-zero (else the PDE is
not of second order). We shall observe how (3.6.1) changes under coordinate
transformation.
Definition 3.6.1. For any PDE of the form (3.6.1) we define its discrimi-
nant as B 2 − AC.
Let T : R2 → R2 be the coordinate transformation with components

T = (w, z), where w, z : R2 → R. We assume that w(x, y), z(x, y) are such
that w, z are both continuous and twice differentiable w.r.t (x, y), and the
Jacobian J of T is non-zero,

wx wy
J = 6= 0.
zx zy
We compute the derivatives of u in the new variable,
ux = uw w x + uz z x ,
uy = uw w y + uz z y ,
uxx = uww wx2 + 2uwz wx zx + uzz zx2 + uw wxx + uz zxx
uyy = uww wy2 + 2uwz wy zy + uzz zy2 + uw wyy + uz zyy
uxy = uww wx wy + uwz (wx zy + wy zx ) + uzz zx zy + uw wxy + uz zxy
Substituting above equations in (3.6.1), we get
a(w, z)uww + 2b(w, z)uwz + c(w, z)uzz = d(w, z, u, uw , uz ).
where D transforms in to d and
a(w, z) = Awx2 + 2Bwx wy + Cwy2 (3.6.2)

b(w, z) = Awx zx + B(wx zy + wy zx ) + Cwy zy (3.6.3)
c(w, z) = Azx2 + 2Bzx zy + Czy2 . (3.6.4)
Note that the coefficients in the new coordinate system satisfy
b2 − ac = (B 2 − AC)J 2 .
Since J 6= 0, we have J 2 > 0. Thus, both b2 − ac and B 2 − AC have the

same sign. Thus, the sign of the discriminant is invariant under coordinate
transformation. All the above arguments can be carried over to quasi-linear
and non-linear PDE.
3.7 Standard or Canonical Forms

The advantage of above classification helps us in reducing a given PDE into
simple forms. Given a PDE, one can compute the sign of the discriminant
and depending on its clasification we can choose a coordinate transformation
(w, z) such that
(i) For hyperbolic, a = c = 0 or b = 0 and a = −c.
(ii) For parabolic, c = b = 0 or a = b = 0. We conveniently choose

c = b = 0 situation so that a 6= 0 (so that division by zero is avoided in
the equation for characteristic curves).
(iii) For elliptic, b = 0 and a = c.

If the given second order PDE (3.6.1) is such that A = C = 0, then
6 0) gives
(3.6.1) is of hyperbolic type and a division by 2B (since B =
uxy = D̃(x, y, u, ux , uy )
where D̃ = D/2B. The above form is the first standard form of second order
hyperbolic equation. If we introduce the linear change of variable X = x + y
and Y = x − y in the first standard form, we get the second standard form
of hyperbolic PDE
uXX − uY Y = D̂(X, Y, u, uX , uY ).
If the given second order PDE (3.6.1) is such that A = B = 0, then
(3.6.1) is of parabolic type and a division by C (since C 6= 0) gives
uyy = D̃(x, y, u, ux , uy )
where D̃ = D/C. The above form is the standard form of second order
parabolic equation.
If the given second order PDE (3.6.1) is such that A = C and B = 0,
then (3.6.1) is of elliptic type and a division by A (since A 6= 0) gives
uxx + uyy = D̃(x, y, u, ux , uy )
where D̃ = D/A. The above form is the standard form of second order
elliptic equation.
Note that the standard forms of the PDE is an expression with no mixed
derivatives.
3.8 Reduction to Standard Form

Consider the second order semi-linear PDE (3.6.1) not in standard form. We
look for transformation w = w(x, y) and z = z(x, y), with non-vanishing
Jacobian, such that the reduced form is the standard form.
If B 2 − AC > 0, we have two characteristics. We are looking for the
coordinate system w and z such that a = c = 0. This implies from equation
(3.6.2) and (3.6.4) that we need to find w and z such that
√
wx −B ± B 2 − AC zx
= = .
wy A zy
Therefore, we need to find w and z such that along the slopes of the charac-
teristic curves, √
dy B ± B 2 − AC −wx
= = .
dx A wy
This means that, using the parametrisation of the characteristic curves,
˙ = 0. Similarly for z. Thus, w and z
wx γ̇1 (s) + wy γ̇2 (s) = 0 and w(s)
are chosen such that they are constant on the characteristic curves.
The characteristic curves are found by solving
√
dy B ± B 2 − AC
=
dx A
and the coordinates are then chosen such that along the characteristic curve
w(x, y) √
2
and z(x, y) = a constant. Note that wx zy − wy zx =
= a constant
2
wy zy A B − AC 6= 0.
Example 3.11. Let us reduce the PDE uxx − c2 uyy = 0 to its canonical form.
Note that A = 1, B = 0, C = −c2 and B 2 − AC = c2 and the equation is
hyperbolic. The characteristic curves are given by the equation
√
dy B ± B 2 − AC
= = ±c.
dx A
Solving we get y ∓ cx = a constant. Thus, w = y + cx and z = y − cx. Now
writing

= c2 (uww − 2uwz + uzz )
= uww + 2uwz + uzz
−c2 uyy = −c2 (uww + 2uwz + uzz )
Substituting into the given PDE, we get
0 = 4c2 uwz
= uwz .
Example 3.12. Let us reduce the PDE uxx − x2 yuyy = 0 given in the region
{(x, y) | x ∈ R, x 6= 0, y > 0} to its canonical form. Note that A = 1, B = 0,
C = −x2 y and B 2 − AC = x2 y. In the given region x2 y > 0, hence the

equation is hyperbolic. The characteristic curves are given by the equation
√
dy B ± B 2 − AC √
= = ±x y.
dx A
√ √
Solving we get x2 /2 ∓ 2 y = a constant. Thus, w = x2 /2 + 2 y and
√
z = x2 /2 − 2 y. Now writing
ux = uw wx + uz zx = x(uw + uz )
1
uy = uw wy + uz zy = √ (uw − uz )
y
= x2 (uww + 2uwz + uzz ) + uw + uz
1 1
= (uww − 2uwz + uzz ) − √ (uw − uz )
y 2y y
x2
−x2 yuyy = −x2 (uww − 2uwz + uzz ) + √ (uw − uz )
2 y
√ √
2 2 y + x2 2 y − x2
0 = 4x uwz + √ uw + √ uz
2 y 2 y
√ √ √
= 8x2 yuwz + (x2 + 2 y)uw + (2 y − x2 )uz .
√
Note that w + z = x2 and w − z = 4 y. Now, substituting x, y in terms of
w, z, we get

2 2 w−z w−z
0 = 2(w − z )uwz + w + z + uw + − w − z uz
2 2

3w + z w + 3z
= uwz + uw − uz .
4(w2 − z 2 ) 4(w2 − z 2 )
Example 3.13. Let us reduce the PDE uxx + uxy − 2uyy + 1 = 0 given in the
region {(x, y) | 0 ≤ x ≤ 1, y > 0} to its canonical form. Note that A = 1,
B = 1/2, C = −2 and B 2 −AC = 9/4 > 0. Hence the equation is hyperbolic.
The characteristic curves are given by the equation
√
dy B ± B 2 − AC 1 3
= = ± = 2 or − 1.
dx A 2 2
Solving we get y − 2x = a constant and y + x = a constant. Thus, w = y − 2x

and z = y + x.
In the parabolic case, B 2 − AC = 0, we have a single characteristic. We
are looking for a coordinate system such that either b = c = 0.
Example 3.14. Let us reduce the PDE e2x uxx + 2ex+y uxy + e2y uyy = 0 to its
canonical form. Note that A = e2x , B = ex+y , C = e2y and B 2 − AC = 0.
The PDE is parabolic. The characteristic curves are given by the equation
dy B ey
= = x.
dx A e
Solving, we get e − e = a constant. Thus, w = e−y − e−x . Now, we
−y −x
choose z such that wx zy − wy zx 6= 0. For instance, z = x is one such choice.

Then
ux = e−x uw + uz
uy = −e−y uw
uxx = e−2x uww + 2e−x uwz + uzz − e−x uw
uyy = e−2y uww + e−y uw
uxy = −e−y (e−x uww − uwz )
ex e−y uzz = (e−y − e−x )uw
Replacing x, y in terms of w, z gives
w
uzz = uw .
1 + wez
1
Example 3.15. Let us reduce the PDE y 2 uxx − 2xyuxy + x2 uyy = xy (y 3 ux +
x3 uy ) to its canonical form. Note that A = y 2 , B = −xy, C = x2 and
B 2 − AC = 0. The PDE is parabolic. The characteristic curves are given by
the equation
dy B −x
= = .
dx A y
Solving, we get x2 + y 2 = a constant. Thus, w = x2 + y 2 . Now, we choose z
such that wx zy − wy zx 6= 0. For instance, z = x is one such choice. Then
ux = 2xuw + uz
uy = 2yuw
In the elliptic case, B 2 − AC < 0, we have no real characteristics. Thus,

we choose w, z to be the real and imaginary part of the solution of the
characteristic equation.
Example 3.16. Let us reduce the PDE x2 uxx + y 2 uyy = 0 given in the region
{(x, y) ∈ R2 | x > 0, y > 0} to its canonical form. Note that A = x2 ,
B = 0, C = y 2 and B 2 − AC = −x2 y 2 < 0. The PDE is elliptic. Solving the
characteristic equation
dy iy
=±
dx x
we get ln x ± i ln y = c. Let w = ln x and z = ln y. Then
ux = uw /x
uy = uz /y
uxx = −uw /x2 + uww /x2
uyy = −uz /y 2 + uzz /y 2
Substituting into the PDE, we get
uww + uzz = uw + uz .
Example 3.17. Let us reduce the PDE uxx +2uxy +5uyy = xux to its canonical
form. Note that A = 1, B = 1, C = 5 and B 2 − AC = −4 < 0. The PDE is
elliptic. The characteristic equation is
dy
= 1 ± 2i.
dx
Solving we get x − y ± i2x = c. Let w = x − y and z = 2x. Then
ux = uw + 2uz
uy = −uw
uxx = uww + 4uwz + 4uzz
uyy = uww
uxy = −(uww + 2uwz )
Substituting into the PDE, we get
uww + uzz = x(uw + 2uz )/4.
Replacing x, y in terms of w, z gives
z
uww + uzz = (uw + 2uz ).
8
Example 3.18. Let us reduce the PDE uxx + uxy + uyy = 0 to its canonical
form. Note that A = 1, B = 1/2, C = 1 and B 2 − AC = −3/4 < 0. The
PDE is elliptic. Solving the characteristic equation
√
dy 1 3
= ±i
dx 2 2
√ √
we get 2y = x ± i 3x + c. Let w = 2y − x and z = 3x.
Exercise 17. Rewrite the PDE in their canonical forms and solve them.
√
(a) uxx + 2 3uxy + uyy = 0
(b) x2 uxx − 2xyuxy + y 2 uyy + xux + yuy = 0
(c) uxx − (2 sin x)uxy − (cos2 x)uyy − (cos x)uy = 0
(d) uxx + 4uxy + 4uyy = 0

Chapter 4
Wave Equation
4.1 One Dimension

The one dimensional wave equation is the first ever partial differential equa-
tion (PDE) to be studied by mankind, introduced in 1752 by d’Alembert as
a model to study vibrating strings. He introduced the one dimensional wave
equation
∂ 2 u(x, t) ∂ 2 u(x, t)
= .
∂t2 ∂x2
d’Alembert used the travelling wave technique to solve the wave equation.
In this chapter we shall explain this technique of d’Alembert and also give
the standing wave technique which motivates the idea of separation of vari-
able and in turn the evolution of Fourier series. The wave equations was
generalised to two and three dimensions by Euler (1759) and D. Bernoulli
(1762), respectively. Note the invariance of the wave equation under the
transformation t 7→ −t, i.e., if u(x, t) is a solution to the wave equation for
t ≥ 0, then û(x, t) := u(x, τ ) is a solution of the wave equation for t < 0 and
τ := −t > 0, because dτ /dt = −1, ut (x, t) = −uτ (x, τ ). This means that
wave equation is reversible in time and do not distinguish between past and
future.
4.1.1 The Vibrating String: Derivation

Let us consider a homogeneous string of length L, stretched along the x-axis,
with one end fixed at x = 0 and the other end fixed at x = L. We assume
that the string is free to move only in the vertical direction. Let ρ > 0 denote
81
CHAPTER 4. WAVE EQUATION 82
the density of the string and T > 0 denote the coefficient of tension of the
string. Let u(x, t) denote the vertical displacement of the string at the point
x and time t.
We shall imagine the string of length L as system of N objects, for N
sufficiently large. Think of N objects sitting on the string L at equidistant
(uniformly distributed). The position of the n-th object on the string is
given by xn = nL/N . One can think of the vibrating string as the harmonic
oscillator of N objects governed by the tension on the string (which behaves
like the spring). Let yn (t) = u(xn , t) denote the displacement of the object xn
at time t. The distance between any two successive objects is h = xn+1 −xn =
L/N . Then mass of each of the N object is mass of the string divided by N .
Since mass of the string is ρ×L, mass of each of the object xn , n = 1, 2, . . . , N ,
is ρh. Thus, by Newton’s second law, ρhyn′′ (t) is same as the force acting on
the n-th object. The force on xn is coming both from left (xn−1 ) and right
(xn+1 ) side. The force from left and right is given as T (yn−1 − yn )/h and
T (yn+1 − yn )/h, respectively. Therefore,
T
ρhyn′′ (t) = {yn+1 (t) + yn−1 (t) − 2yn (t)}
h
T
= {u(xn + h, t) + u(xn − h, t) − 2u(xn , t)}
h
′′ T u(xn + h, t) + u(xn − h, t) − 2u(xn , t)
yn (t) =
ρ h2
Note that assuming u is twice differentiable w.r.t the x variable, the term
on RHS is same as

T 1 u(xn + h, t) − u(xn , t) u(xn − h, t) − u(xn , t)
+
ρh h h
which converges to the second partial derivative of u w.r.t x as h → 0. The
h → 0 is the limit case of the N objects we started with. Therefore the
vibrating string system is governed by the equation
∂ 2u T ∂ 2u
=
∂t2 ρ ∂x2
where T is the tension and ρ is the density of the string. Equivalently,
∂ 2u 2
2∂ u
= c (4.1.1)
∂t2 ∂x2
where c2 = T /ρ, c > 0, on x ∈ (0, L) and t > 0.
Exercise 18. Show that the wave equation (4.1.1), derived above can be
written as
uzz = uww in (w, z) ∈ (0, L) × (0, ∞).
under a new coordinate system (w, z). One may, in fact choose coordinate
such that the string is fixed between (0, π).
Proof. Set w = x/a and z = t/b, where a and b will be chosen appropriately.
Then, wx = 1/a and zt = 1/b. Therefore, ux = uw /a, ut = uz /b, a2 uxx = uww
and b2 utt = uzz . Choosing a = 1 and b = 1/c. Choosing a = L/π and
b = L/cπ makes the domain (0, π).
4.1.2 Travelling Waves

Consider the wave equation utt = c2 uxx on R × (0, ∞), describing the vibra-
tion of an infinite string. We have already seen in Chapter 3 that the equation
is hyperbolic and has the two characteristics x ± ct= a constant. Introduce
the new coordinates w = x + ct, z = x − ct and set u(w, z) = u(x, t). Thus,
we have the following relations, using chain rule:
ux = u w w x + u z z x = u w + u z
ut = uw wt + uz zt = c(uw − uz )
uxx = uww + 2uzw + uzz
utt = c2 (uww − 2uzw + uzz )
In the new coordinates, the wave equation satisfies uwz = 0. Integrating1

this twice, we have u(w, z) = F (w) + G(z), for some arbitrary functions F
and G. Thus, u(x, t) = F (x + ct) + G(x − ct) is a general solution of the
wave equation.
Consider the case where G is chosen to be zero function. Then u(x, t) =
F (x+ct) solves the wave equation. At t = 0, the solution is simply the graph
of F and at t = t0 the solution is the graph of F with origin translated to the
left by ct0 . Similarly, choosing F = 0 and G = F , we have u(x, t) = F (x−ct)
also solves wave equation and at time t is the translation to the right of the
graph of F by ct. This motivates the name “travelling waves” and “wave
equation”. The graph of F is shifted to right or left with a speed of c.
1
We are assuming the function is integrable, which may be false
Now that we have derived the general form of the solution of wave equa-
tion, we return to understand the physical system of a vibrating infinite
string. The initial shape (position at initial time t = 0) of the string is given
as u(x, 0) = g(x), where the graph of g on R2 describes the shape of the
string. Since we need one more data to identify the arbitrary functions, we
also prescribe the initial velocity of the string, ut (x, 0) = h(x).
Another interesting property that follows from the general solution is
that for any four points A, B, C and D that form a rectangle bounded by
characteristic curves in R × R+ then u(A) + u(C) = u(B) + u(D) because
u(A) = F (α) + G(β), u(C) = F (γ) + G(δ), u(B) = F (α) + G(δ) and u(D) =
F (γ) + G(β).
Theorem 4.1.1. Given g ∈ C 2 (R) and h ∈ C 1 (R), there is a unique C 2
solution u of the Cauchy initial value problem (IVP) of the wave equation,

 utt (x, t) − c2 uxx (x, t) = 0 in R × (0, ∞)
u(x, 0) = g(x) in R (4.1.2)

ut (x, 0) = h(x) in R,
which is given by the d’Alembert’s formula

Z x+ct
1 1
u(x, t) = (g(x + ct) + g(x − ct)) + h(y) dy. (4.1.3)
2 2c x−ct
Proof. The general solution is u(x, t) = F (x + ct) + G(x − ct) with F, G ∈

C 2 (R). Using the initial position we get
F (x) + G(x) = g(x).
Thus, g should be C 2 (R). Now, ut (x, t) = c (F ′ (w) − G′ (z)) and putting

t = 0, we get
1
F ′ (x) − G′ (x) = h(x).
c
Thus, h should be C (R). Now solving for F ′ and G′ , we get 2F ′ (x) =
1
g ′ (x) + h(x)/c. Similarly, 2G′ (x) = g ′ (x) − h(x)/c. Integrating2 both these
equations, we get
Z
1 1 x
F (x) = g(x) + h(y) dy + c1
2 c 0
2
assuming they are integrable and the integral of their derivatives is itself
and Z x
1 1
G(x) = g(x) − h(y) dy + c2 .
2 c 0
Since F (x) + G(x) = g(x), we get c1 + c2 = 0. Therefore, the solution to the
wave equation is given by (4.1.3).
Aliter. Let us derive the d’Alembert’s formula in an alternate way. Note
that the wave equation can be factored as

∂ ∂ ∂ ∂
+c −c u = utt − c2 uxx = 0.
∂t ∂x ∂t ∂x
∂ ∂

We set v(x, t) = ∂t − c ∂x u(x, t) and hence
vt (x, t) + cvx (x, t) = 0 in R × (0, ∞).
Notice that the above first order PDE obtained is in the form of homogeneous
linear transport equation (cf. (2.2.1)), which we have already solved. Hence,
for some smooth function φ,
v(x, t) = φ(x − ct)
and φ(x) := v(x, 0). Using v in the original equation, we get the inhomoge-
neous transport equation,
ut (x, t) − cux (x, t) = φ(x − ct).
Recall the formula for inhomogenoeus transport equation (cf. (2.2.2))
Z t
u(x, t) = g(x − at) + φ(x − a(t − s), s) ds.
0
Since u(x, 0) = g(x) and a = −c, in our case the solution reduces to,
Z t
u(x, t) = g(x + ct) + φ(x + c(t − s) − cs) ds
0
Z t
= g(x + ct) + φ(x + ct − 2cs) ds
0
Z
−1 x−ct
= g(x + ct) + φ(y) dy
2c x+ct
Z
1 x+ct
= g(x + ct) + φ(y) dy.
2c x−ct
But φ(x) = v(x, 0) = ut (x, 0) − cux (x, 0) = h(x) − cg ′ (x) and substituting
this in the formula for u, we get
Z
1 x+ct
u(x, t) = g(x + ct) + (h(y) − cg ′ (y)) dy
2c x−ct
1
= g(x + ct) + (g(x − ct) − g(x + ct))
Z x+ct 2
1
+ h(y) dy
2c x−ct
Z
1 1 x+ct
= (g(x − ct) + g(x + ct)) + h(y) dy
2 2c x−ct
For c = 1, the d’Alembert’s formula takes the form

Z
1 1 x+t
u(x, t) = (g(x − t) + g(x + t)) + h(y) dy.
2 2 x−t
A useful observation from the d’Alembert’s formula is that the regularity of
u is same as the regularity of its initial value g.
Theorem 4.1.2 (Dirichlet Condition). Given g ∈ C 2 [0, ∞), h ∈ C 1 [0, ∞)
and φ ∈ C 2 (0, ∞), there is a unique C 2 solution u of the homogeneous Cauchy
initial value problem (IVP) of the wave equation,


 utt (x, t) − c2 uxx (x, t) = 0 in (0, ∞) × (0, ∞)

u(x, 0) = g(x) in [0, ∞)
(4.1.4)

 ut (x, 0) = h(x) in [0, ∞)

u(0, t) = φ(t) in (0, ∞),
where φ, g, h satisfies the compatibility condition
g(0) = φ(0), g ′′ (0) = φ′′ (0), h(0) = φ′ (0).
Proof. We first consider case φ ≡ 0, homogeneous Dirichlet conditions. We

extend g as an odd function on (−∞, ∞) by setting g(−x) = −g(x) for
x ∈ [0, ∞). Therefore, g is an odd function in R. Then, we have a unique
solution u ∈ C 2 solving the Cauchy problem in R × (0, ∞). But note that
v(x, t) = −u(−x, t) is also a solution to the same Cauchy problem. Thus,
u(x, t) = −u(−x, t) for all (x, t) ∈ R × R+ and u(0, t) = 0.
Now, we consider the case when φ is a non-zero function. The line ct = x

divides the quadrant in to two regions Ω1 and Ω2 defined as
Ω1 = {(x, t) | x > ct}
and
Ω2 = {(x, t) | x < ct}.
For (x, t) ∈ Ω1 , the solution
Z x+ct
1 1
u1 (x, t) = (g(x − ct) + g(x + ct)) + h(y) dy.
2 2c x−ct
On the line x = ct, we get

Z 2x
1 1
χ(x) := u1 (x, x/c) = (g(0) + g(2x)) + h(y) dy.
2 2c 0
Let u2 be the solution in Ω2 of


 utt (x, t) − c2 uxx (x, t) = 0 in Ω2
u(x, x/c) = χ(x) in {x = ct}

u(0, t) = φ(t) in (0, ∞).
Fix A := (x, t) ∈ Ω2 . One of the characteristic curve through A intersects

t-axis at B := (0, t − x/c). The other characteristic curve intersects the
line ct = x at C := (1/2(ct + x, t + x/c). The characteristic curve through
B intersects ct = x at D := 1/2(ct − x, t − x/c). The four points form a
parallelogram in Ω2 . Therefore, we know that
u2 (x, t) + u2 (1/2(ct − x, t − x/c)) = u2 (0, t − x/c) + u2 (1/2(ct + x, t + x/c))
and, hence,
u2 (x, t) = φ(t − x/c) + χ(1/2(ct + x)) − χ(1/2(ct − x))

Z
1 1 ct+x
= φ(t − x/c) + (g(0) + g(ct + x)) + h(y) dy
2 2c 0
Z
1 1 ct−x
− (g(0) + g(ct − x)) − h(y) dy
2 2c 0
Z
1 1 ct+x
= φ(t − x/c) + (g(ct + x) − g(ct − x)) + h(y) dy.
2 2c ct−x
By setting (
u1 (x, t) (x, t) ∈ Ω1
u(x, t) =
u2 (x, t) (x, t) ∈ Ω2
and the fact that alll derivatives of u are continuous along ct = x line, due
to the compatibility condition implies that u is a solution (4.1.4).
Corollary 4.1.3 (Dirichlet Condition). Given g ∈ C 2 [0, L], h ∈ C 1 [0, L] and
φ, ψ ∈ C 2 (0, ∞), there is a unique C 2 solution u of the homogeneous Cauchy
initial value problem (IVP) of the wave equation,
 2
 utt (x, t) − c uxx (x, t) = 0


in (0, L) × (0, ∞)

 u(x, 0) = g(x) in [0, L]
ut (x, 0) = h(x) in [0, L] (4.1.5)



 u(0, t) = φ(t) in (0, ∞)

u(L, t) = ψ(t) in (0, ∞),
where φ, ψ, g, h satisfies the compatibility condition
g(0) = φ(0), g ′′ (0) = φ′′ (0), h(0) = φ′ (0)
and
g(L) = ψ(0), g ′′ (L) = ψ ′′ (0), h(L) = ψ ′ (0).
Proof. We first consider case φ = ψ ≡ 0, homogeneous Dirichlet conditions.
We extend g as an odd function on [−L, L] by setting g(−x) = −g(x) for
x ∈ [0, L]. Therefore, g is an odd 2L-periodic function in R. Similarly,
we extend h and we have g ∈ C 2 (R) and h ∈ C 1 (R). Then, we have a
unique solution u ∈ C 2 solving the Cauchy problem in R × (0, ∞). But
note that v(x, t) = −u(−x, t) is also a solution to the same Cauchy problem.
Thus, u(x, t) = −u(−x, t) for all (x, t) ∈ R × R+ and u(0, t) = 0. Similarly,
w(x, t) = −u(2L − x, t) is also a solution and, hence, u(x, t) = −u(2L − x, t)
which implies u(L, t) = 0.
Consider the lines ct = x and ct = −x + cL, then we will obtain u in the
four regions as u1 , u2 , u3 and u4 . Then follow the idea similar to proof in
above theorem.
4.1.3 Domain of Dependence and Influence

Note that the solution u(x, t) depends only on the interval [x − ct, x + ct]
because g takes values only on the end-points of this interval and h takes
values between this interval. The interval [x − ct, x + ct] is called the domain
of dependence. Thus, the region of R × (0, ∞) on which the value of u(x, t)
depends forms a triangle with base [x − ct, x + ct] and vertex at (x, t). The
domain of dependence of (x, t) is marked in x-axis by the characteristic curves
passing through (x, t).
Given a point p on the x-axis what values of u on (x, t) will depend on
the value of g(p) and h(p). This region turns out to be a cone with vertex
at p and is called the domain of influence. The domain of influence is the
region bounded by the characteristic curves passing through p.
If the initial data g and h are supported in the interval Bx0 (R) then the
solution u at (x, t) is supported in the region Bx0 (R + ct). Consequently, if g
and h have compact support then the solution u has compact support in R
for all time t > 0. This phenomenon is called the finite speed of propagation.
Theorem 4.1.4 (Inhomogeneous Wave Equation). Given g ∈ C 2 (R), h ∈

C 1 (R) and f ∈ C 1 (R × [0, ∞), there is a unique C 2 solution u of the inho-
mogeneous Cauchy initial value problem (IVP) of the wave equation,

 utt (x, t) − c2 uxx (x, t) = f (x, t) in R × (0, ∞)
u(x, 0) = g(x) in R (4.1.6)

ut (x, 0) = h(x) in R,
given by the formula

"Z Z tZ #
x+ct x+c(t−s)
1 1
[g(x + ct) + g(x − ct)] + h(y) dy + f (y, s) dy ds .
2 2c x−ct 0 x−c(t−s)
Proof. Fix (x, t) ∈ R × R+ . Consider the open triangle in R × R+ with

vertices (x, t), (x − ct, 0) and (x + ct, 0), and denote it by T (x, t). Thus,
T (x, t) := {(y, s) ∈ R × R+ | |y − x| < c(t − s)}.
The boundary of the triangle ∂T (x, t) consists of three parts
T0 := {(y, 0) | x − ct < y < x + ct},
T+ := {(y, s) ∈ R × (0, t) | cs + y = x + ct}

and
T− := {(y, s) ∈ R × (0, t) | cs − y = −x + ct}.
The unit outward normal at each point of the boundary ∂T (x, t) is given by
ν = (ν1 , ν2 ) defined as

(0, −1)
 (y, s) ∈ T0
1
ν(y, s) = √1+c2 (1, c) (y, s) ∈ T+

√ 1
1+c2
(−1, c) (y, s) ∈ T− .
Therefore, by Gauss divergence theorem (cf. (D.0.1))

Z Z

f ((y, s) dy ds = utt (y, s) − c2 uxx (y, s) dy ds
T (x,t)
ZT (x,t)

= ut ν2 − c2 ux ν1 dσ
Z∂T (x,t) Z
2

= ut ν2 − c ux ν1 dσ + ut ν2 − c2 ux ν1 dσ
T0 T+
Z

+ ut ν2 − c2 ux ν1 dσ
T−
Z x+ct Z
c
= − ut (y, 0) dy + √ [ut − cux ] dσ
x−ct 1 + c2 T+
Z
c
+√ [ut + cux ] dσ
1 + c2 T−
Note that the second and third integral are just the direction derivatives of
u along the direction (1, c) and (−1, c) in the line T+ and T− , respectively.
Therefore, we have
Z Z x+ct
f ((y, s) dy ds = − h(y) dy + cu(x, t) − cu(x + ct, 0)
T (x,t) x−ct
+c[u(x, t) − u(x − ct, 0)]
1
u(x, t) = [g(x + ct) + g(x − ct)]
2 " #
Z x+ct Z t Z x+c(t−s)
1
+ h(y) dy + f (y, s) dy ds .
2c x−ct 0 x−c(t−s)
Aliter. We introduce a new function v defined as v(x, t) = ut (x, t) and rewrite

(4.1.6) as
U ′ (x, t) + AU (x, t) = F (x, t)
where U = (u, v), F = (0, f ) and

0 −1
A= ∂ 2
c2 ∂x 2 0
with the initial condition G(x) := U (x, 0) = (g(x), h(x)). The solution
U (x, t) is given as (cf. Appendix G)
Z t
U (x, t) = S(t)G(x) + S(t − s)F (s) ds
0
where S(t) is a solution operator of the homoegeneous system of first order

PDE. Therefore, by d’Alembert’s formula,
1 1
R x+ct
[g(x + ct) + g(x − ct)] + h(y) dy
S(t)(g, h) = c
2 2c x−ct
2
[g ′ (x + ct) − g ′ (x − ct)] + 21 [h(x + ct) + h(x − ct)]
and, hence, u(x, t) has the required represenation.
4.1.4 Standing Waves: Separation of Variable

Recall the set-up of the vibrating string given by the equation utt = uxx ,
we have normalised the constant c. Initially at time t, let us say the string
has the shape of the graph of v, i.e., u(x, 0) = v(x). The snapshot of the
vibrating string at each time are called the “standing waves”. The shape of
the string at time t0 can be thought of as some factor (depending on time)
of v. This observation motivates the idea of “separation of variable”, i.e.,
u(x, t) = v(x)w(t), where w(t) is the factor depending on time, which scales
v at time t to fit with the shape of u(x, t).
The fact that endpoints are fixed is given by the boundary condition
u(0, t) = u(L, t) = 0.
We are also given the initial position u(x, 0) = g(x) (at time t = 0) and initial
velocity of the string at time t = 0, ut (x, 0) = h(x). Given g, h : [0, L] → R
such that g(0) = g(L) = 0 and h(0) = h(L), we need to solve the initial value
problem (4.1.5) with φ = ψ ≡ 0.
Let us seek for solutions u(x, t) whose variables can be separated. Let
u(x, t) = v(x)w(t). Differentiating and substituting in the wave equation, we
get
v(x)w′′ (t) = c2 v ′′ (x)w(t)
Hence
w′′ (t) v ′′ (x)
= .
c2 w(t) v(x)
Since RHS is a function of x and LHS is a function t, they must equal a
constant, say λ. Thus,
v ′′ (x) w′′ (t)
= 2 = λ.
v(x) c w(t)
Using the boundary condition u(0, t) = u(L, t) = 0, we get
v(0)w(t) = v(L)w(t) = 0.
If w ≡ 0, then u ≡ 0 and this cannot be a solution to (4.1.5). Hence, w 6≡ 0

and v(0) = v(L) = 0. Thus, we need to solve the eigen value problem for the
second order differential operator.
′′
v (x) = λv(x), x ∈ (0, L)
v(0) = v(L) = 0,
Note that the λ can be either zero, positive or negative. If λ = 0, then

′′
v = 0 and the general solution is v(x) = αx + β, for some constants α and
β. Since v(0) = 0, we get β = 0, and v(L) = 0 and L 6= 0 implies that α = 0.
Thus, v ≡ 0 and hence u ≡ √ 0. But, this
√
cannot be a solution to (4.1.5).
λx − λx
If λ > 0, then v(x) = αe + βe . Equivalently,
√ √
v(x) = c1 cosh( λx) + c2 sinh( λx)
such that α = (c1 + c2 )/2 and β = (c1 − c2 )/2. Using the boundary condition
v(0) = 0, we get c1 = 0 and hence
√
v(x) = c2 sinh( λx).
√
Now using v(L) = 0, we have c2 sinh λL = 0. Thus, c2 = 0 and v(x) = 0.
We have seen this cannot be a solution.
√
Finally, if λ < 0, then set ω = −λ. We need to solve the simple
harmonic oscillator problem
′′
v (x) + ω 2 v(x) = 0 x ∈ (0, L)
v(0) = v(L) = 0.
The general solution is
v(x) = α cos(ωx) + β sin(ωx).
Using v(0) = 0, we get α = 0 and hence v(x) = β sin(ωx). Now using

v(L) = 0, we have β sin ωL = 0. Thus, either β = 0 or sin ωL = 0. But
β = 0 does not yield a solution. Hence ωL = kπ or ω = kπ/L, for all non-
zero k ∈ Z. Since ω > 0, we can consider only k ∈ N. Hence, for each k ∈ N,
there is a solution (vk , λk ) for the eigen value problem with

kπx
vk (x) = βk sin ,
L
for some constant bk and λk = −(kπ/L)2 . It now remains to solve w for each
of these λk . For each k ∈ N, we solve for wk in the ODE
wk′′ (t) + (ckπ/L)2 wk (t) = 0.

ckπt ckπt
wk (t) = ak cos + bk sin .
L L
For each k ∈ N, we have

ckπt ckπt kπx
uk (x, t) = ak cos + bk sin sin
L L L
for some constants ak and bk . The situation corresponding to k = 1 is called
the fundamental mode and the frequency of the fundamental mode is
p
1 cπ c T /ρ
= = .
2π L 2L 2L
The frequency of higher modes are integer multiples of the fundamental fre-
quency.
The general solution of (4.1.5), by principle of superposition, is

X∞
ckπt ckπt kπx
u(x, t) = ak cos + bk sin sin .
k=1
L L L
Note that the solution is expressed as series, which raises the question of
convergence of the series. Another concern is whether all solutions of (4.1.5)
have this form. We ignore these two concerns at this moment.
Since we know the initial position of the string as the graph of g, we get
X∞
kπx
g(x) = u(x, 0) = ak sin .
k=1
L
This expression is again troubling and rises the question: Can any arbitrary
function g be expressed as an infinite sum of trigonometric functions? An-
swering this question led to the study of “Fourier series”. Let us also, as
usual, ignore this concern for time being. Then, can we find the the con-
lπx
stants ak with knowledge of g. By multiplying sin L both sides of the
expression of g and integrating from 0 to L, we get
Z L Z L "X ∞ #
lπx kπx lπx
g(x) sin dx = ak sin sin dx
0 L 0 k=1
L L
X∞ Z L
kπx lπx
= ak sin sin dx
k=1 0 L L
Therefore, the constants ak are given as
Z
2 L kπx
ak = g(x) sin .
L 0 L
Finally, by differentiating u w.r.t t, we get
X∞
ckπ ckπt ckπt kπx
ut (x, t) = bk cos − ak sin sin .
k=1
L L L L
Employing similar arguments and using ut (x, 0) = h(x), we get
X∞
bk kcπ kπx
h(x) = ut (x, 0) = sin
k=1
L L
and hence Z
L
2 kπx
bk = h(x) sin .
kcπ 0 L
4.2 Higher Dimensions

We shall denote ∆ as the Laplacian w.r.t. the space variable and the wave
equation is denoted as u := utt − c2 ∆u. The Cauchy initial value problem
in higher dimensions is

 u = f (x, t) in Rn × (0, ∞)
u(x, 0) = g(x) in Rn (4.2.1)

ut (x, 0) = h(x) in Rn .
Due to the linearity of the wave operator , any solution u = v +w +z where

v, w and z are, respectively, solutions of

 v = 0 in Rn × (0, ∞)
v(x, 0) = g(x) in Rn (4.2.2)

vt (x, 0) = 0 in Rn ,

 w = 0 in Rn × (0, ∞)
w(x, 0) = 0 in Rn (4.2.3)

wt (x, 0) = h(x) in Rn
and 
 z = f (x, t) in Rn × (0, ∞)
z(x, 0) = 0 in Rn (4.2.4)

zt (x, 0) = 0 in Rn .
Theorem 4.2.1 (Duhamel’s Principle). Let wh be a solution to (4.2.3). Then
v(x, t) = wtg (x, t)
and Z t
z(x, t) = wfs (x, t − s) ds,
0
where fs = f (·, s), are solutions to (4.2.2) and (4.2.4).
Proof. Since wg solves (4.2.3) with h = g, we have
∂
v = wtg (x, t) = (wg ) = 0
∂t
on Rn × (0, ∞). Further, v(x, 0) = wtg (x, 0) = g(x) in Rn and vt (x, 0) =
wttg (x, 0) = c2 ∆wg (x, 0) = 0. Thus, v solves (4.2.2). Now, let us denote
w(x, t) = wfs (x, t−s), for 0 < s < t, which satisfies (4.2.3) with h(·) = f (c,̇s).
Then,
Z t Z t
fs
fs
zt (x, t) = w (x, 0) + wt (x, t − s) ds = wtfs (x, t − s) ds
0 0
and
Z t
ztt (x, t) = wtfs (x, 0) + wttfs (x, t − s) ds
0
Z t
2
= f (x, t) + c ∆ wfs (x, t − s) ds
0
= f (x, t) + c2 ∆z.
Therefore, z solves (4.2.4).

The Duhamel’s principle can be viewed as a generalization of the method
of variations of constants in ODE (cf. Appendix G). Owing to the above
theorem it is enough solve for w in (4.2.3). To do so, we shall employ the
method of spherical means which reduces such problem to one dimensional
framework.
4.2.1 Spherical Means

More generally, in this section we solve for v + w which is the solution to the
wave equation 
 u = 0 in Rn × (0, ∞)
u(x, 0) = g(x) in Rn (4.2.5)
 n
ut (x, 0) = h(x) in R .
For a fixed x ∈ Rn and t ∈ (0, ∞), the spherical mean of a C 2 function
u(x, t) is given as
Z
1
M (x; r, t) := u(y, t) dσy
ωn rn−1 Sr (x)
where ωn is the surface area of the unit sphere in Rn . Equivalently, after

setting z = (y − x)/r,
Z
1
M (x; r, t) := u(x + rz, t) dσz .
ωn S1 (0)
If we take the above form of M (x; r, t) as a definition, we observe that it is

valid for all r ∈ R and M (x; 0, t) = u(x, t). Then
Z Xn
1
Mr (x; r, t) = uy (x + rz, t)zi dσz
ωn S1 (0) i=1 i
Z
1
= ∇y u(y, t) · z dσy
ωn rn−1 Sr (x)
Z
1
= ∆y u(y, t) dy.
ωn rn−1 Br (x)
If u(x, t) is a solution to (4.2.5) then
Z
n−1 1
r Mr (x; r, t) = 2 utt (y, t) dy
c ωn Br (x)
Z rZ
1
= 2 utt (y, t) dσy ds
c ωn 0 Ss (x)
Z
d n−1 1
(r Mr (x; r, t)) = 2 utt (y, t) dσy
dr c ωn Sr (x)
Z
rn−1 ∂ 2 1 rn−1
= ∇ y u(y, t) dσ y = Mtt .
c2 ∂t2 ωn rn−1 Sr (x) c2
Thus, the spherical means satisfies the one space variable PDE
d n−1
r1−n (r Mr ) = c−2 Mtt
dr
or
n−1
Mrr + Mr = c−2 Mtt ,
r
which is called the Euler-Poisson-Darboux equation. Also, if u(x, t) is a
solution to (4.2.5) then using the initial condition, we get
Z
1
G(x; r) := M (x; r, 0) = g(x + rz) dσz
ωn S1 (0)
and Z
1
H(x; r) := Mt (x; r, 0) = h(x + rz) dσz .
ωn S1 (0)
In the following section we shall illustrate solving the Euler-Poisson-
Darboux equation in three and two dimensions. For higher dimensions solv-
ing E-P-D equation is tedious.
Theorem 4.2.2 (Three dimensions). Given g ∈ C 3 (R3 ) and h ∈ C 2 (R3 )

there exists a unique solution u ∈ C 2 (R3 × [0, ∞) of (4.2.5) given by the
Poisson’s formula
Z Z
1 ∂ 1 1
u(x, t) = g(y) dσy + h(y) dσy .
4πc2 ∂t t Sct (x) t Sct (x)
Proof. For the Euclidean dimension three, the Euler-Poisson-Darboux equa-
tion becomes
c2 (rM )rr = (rM )tt .
Thus, rM is a solution to the one dimensional Cauchy problem
 2
 c (rM )rr (x; r, t) = (rM )tt (x; r, t) in (0, ∞) × (0, ∞)
rM (x; r, 0) = rG(x; r) in (0, ∞)

rMt (x; r, 0) = rH(x; r) in (0, ∞)
and is given by the d’Alembert formula
Z r+ct
1 1
M (x; r, t) = ((r − ct)G(r − ct) + (r + ct)G(r + ct))+ yH(y) dy,
2r 2rc r−ct
as long as, the domain of dependence [r − ct, r + ct] is in (0, ∞). Otherwise,
we extend G and H to entire R as


G(x; r) r>0
G̃(x; r) = g(x) r=0


G(x; −r) r < 0
and 

H(x; r) r>0
H̃(x; r) = h(x) r=0


H(x; −r) r < 0.
Invoking Lemma 4.2.3, we get M̃ (x; r, t) as
Z r+ct
1 1
(r − ct)G̃(r − ct) + (r + ct)G̃(r + ct) + y H̃(y) dy
2r 2rc r−ct
Since G̃ and H̃ are even functions, we have

Z −(r−ct)
yH(y) dy = 0
r−ct
and, therefore M̃ (x; r, t)

Z
1 1 r+ct
(r + ct)G̃(ct + r) − (ct − r)G̃(ct − r) + y H̃(y) dy.
2r 2rc ct−r
Since
u(x, t) = lim+ M̃ (x; r, t),
r→0
by L’Hospital’s rule, we get
d
u(x, t) = (tG(ct)) + tH(ct) = ctG′ (ct) + G(ct) + tH(ct).
dt
Lemma 4.2.3. rG̃, rH̃ ∈ C 2 (R).
Proof. It follows from the definition that
lim G(x; r) = g(x) and lim H(x; r) = h(x).

r→0+ r→0+
Thus, rG̃, rH̃ ∈ C(R). Since G, H ∈ C 1 (R), rG̃, rH̃ ∈ C 1 (R). Further, since
G′′ and H ′′ are bounded as r → 0+, rG̃, rH̃ ∈ C 2 (R).
The domain of dependence of (x, t0 ) for the three dimensional wave equa-
tion is the boundary of the three dimensional sphere with radius ct.
The Hadamard’s method of descent is the technique of finding a solu-
tion of the two dimensional wave equation using the three dimensional wave
equation.
Theorem 4.2.4 (Method of Descent). Given g ∈ C 3 (R2 ) and h ∈ C 2 (R2 )

there exists a unique solution u ∈ C 2 (R2 × [0, ∞) of (4.2.5) given by the
Poisson’s formula
" Z ! Z #
1 ∂ g(ξ, η) h(ξ, η)
p dξ dη + 2 2 2
dξ dη
2πc ∂t Bct (x,y) c 2 t 2 − ρ2 Bct (x,y) c t − ρ
where p
ρ= (ξ − x)2 + (η − y)2 .
Proof. Let v be a solution of (4.2.5) in two dimensions with g ∈ C 3 (R2 ) and

h ∈ C 2 (R2 ). Then
u(x, y, z, t) := v(x, y, t)
is solution to (4.2.5) in three dimensions where g and h are given independent
of z. Since u(x, y, z, t) = u(x, y, 0, t)+zuz (x, y, εz, t) for 0 < ε < 1 and uz = 0,
we have v(x, y, t) = u(x, y, 0, t). Therefore, using the poisson formula in three
dimensions, we get
Z Z
1 ∂ 1 1
v(x, y, t) = g(ξ, η, ζ) dσ + h(ξ, η, ζ) dσ .
4πc2 ∂t t Sct (x,y,0) t Sct (x,y,0)
Recall that g and h are independent of ζ, therefore g(ξ, η, ζ) = g(ξ, η) and

h(ξ, η, ζ) = h(ξ, η). The sphere has the equation (ξ −x)2 +(η−y)2 +ζ 2 = c2 t2 .
The exterior normal is given as

∇S ξ−x η−y ζ
ν= = , ,
|∇S| ct ct ct
and the surface element dσ is given as dσ = dξdη |γ13 | where

p
c2 t2 − (ξ − x)2 − (η − y)2
γ3 = ±
ct
with the positive sign applying when ζ > 0 and negative sign applying when
ζ < 0. Set p
ρ = (ξ − x)2 + (η − y)2 .
In the two dimensions, the domain of dependence is the entire disk

Bct0 (x0 , y0 ) in contrast to three dimensions which had only the boundary
of the sphere as domain of dependence.
4.2.2 Odd Dimension

One can copy the idea of three dimension to any odd dimension, if we rewrite
the Euler-Poisson-Darboux equation in approriate form.
Exercise 19. If n is odd, show that the correct form of M that satisfies the
one dimensional wave equation is
n−3
1 ∂ 2
(rn−2 M (r, t)).
r ∂r
For instance, when n = 5, r2 Mr + 3rM satisfies the one dimensional wave

equation.
We have already noted that the solution at a given point is determined by

the value of intial data in a subset of the initial hypersurface. Consequently,
changing initial value outside the domain of dependence does not change
values of solutions.
Also, it takes time for the initial data to make influence. Suppose g and
h have their support in Br (x0 ). Then the support of u(·, t) is contained in
∪y∈Br (x0 ) Bt (y) = Br+ct (x0 ). The support of u spreads at a finite speed and
is called the finite speed propagation.
4.2.3 Inhomogeneous Wave equation

We have already derived in Theorem 4.2.1 the formula for inhomogeneous
equation (4.2.4).
Theorem 4.2.5. For any f ∈ C 1 , the solution u(x, t) of (4.2.4) is given as

 R t R x+c(t−s)
 1
 f (y, s)dy ds n=1

 2c R0 x−c(t−s)
1 t R f (y,s)
u(x, t) = 4πc √ dy ds n=2

 0 Bc(t−s) (x) c2 (t−s)2 −r 2

 1 R f (y,t−r/c)
4πc2 Bct (x) r
dy n = 3.
Proof. The Poisson’s formula corresponding to the three dimension case gives
the formula for
Z
fs 1
w (x, t − s) = 2
f (y, s) dσy .
4πc (t − s) Sc(t−s) (x)
and
Z t Z tZ
fs 1 f (y, s)
u(x, t) = w (x, t − s) ds = 2
dσy ds
0 4πc 0 Sc(t−s) (x) t − s
Z ct Z
1 f (y, t − τ /c)
= 2
dσy dτ [setting τ = c(t − s)]
4πc 0 Sτ (x) τ
Z
1 f (y, t − r/c)
= 2
dy
4πc Bct (x) r
where r = |x − y|. Similarly, one can derive the formulae for one and two
dimensions.
Note that in the three dimensional case the integrand is not taken at time
t, but at an earlier time t − r/c. Thus, the integrand in this case is called
retarded potential.
Example 4.1. Consider the wave equation

 utt (x, t) − c2 uxx (x, t) = sin 3x in (0, π) × (0, ∞)
u(0, t) = u(π, t) = 0 in (0, ∞)

u(x, 0) = ut (x, 0) = 0 in (0, π).
We look for the solution of the homogeneous wave equation


 wtt (x, t) − c2 wxx (x, t) = 0 in (0, π) × (0, ∞)

w(0, t) = w(π, t) = 0 in (0, ∞)

 w(x, 0) = 0 in (0, π)

wt (x, 0) = sin 3x in (0, π).
By separation of variable technique, we know that the general solution of w
is ∞
X
w(x, t) = [ak cos(kct) + bk sin(kct)] sin(kx)
k=1
and ∞
X
w(x, 0) = ak sin(kx) = 0.
k=1
Thus, ak = 0, for all k ∈ N. Also,
∞
X
wt (x, 0) = bk ck sin(kx) = sin 3x.
k=1
Hence, bk ’s are all zeroes, except for k = 3 and b3 = 1/3c. Thus,

1
w(x, t) = sin(3ct) sin(3x)
3c
and
Z t Z
1 t
u(x, t) = w(x, t − s) ds = sin(3c(t − s)) sin 3x ds
0 3c 0
Z t
sin 3x t sin 3x cos(3c(t − s))
= sin(3c(t − s)) ds =
3c 0 3c 3c 0
sin 3x
= (1 − cos 3ct) .
9c2
4.3 Eigenvalue Problem of Laplacian

The separation of variable technique can be used for studying wave equation
on 2D Rectangle and 2D Disk etc. This leads to studying the eigen value
problem of the Laplacian. For a given open bounded subset Ω ⊂ R2 , the
Dirichlet eigenvalue problem,

−∆u(x, y) = λu(x, y) (x, y) ∈ Ω
u(x, y) = 0 (x, y) ∈ ∂Ω.
Note that, for all λ ∈ R, zero is a trivial solution of the Laplacian. Thus,
we are interested in non-zero λ’s for which the Laplacian has non-trivial
solutions. Such an λ is called the eigenvalue and corresponding solution uλ
is called the eigen function.
Note that if uλ is an eigen function corresponding to λ, then αuλ , for all
α ∈ R, is also an eigen function corresponding to λ. Let W be the real vector
space of all u : Ω → R continuous (smooth, as required) functions such that
u(x, y) = 0 on ∂Ω. For each eigenvalue λ of the Laplacian, we define the
subspace of W as
Wλ = {u ∈ W | u solves Dirichlet EVP for given λ}.
Theorem 4.3.1. There exists an increasing sequence of positive numbers

0 < λ1 < λ2 < λ3 < . . . < λn < . . . with λn → ∞ which are eigenvalues of
the Laplacian and Wn = Wλn is finite dimensional. Conversely, any solution
u of the Laplacian is in Wn , for some n.
Though the above theorem assures the existence of eigenvalues for Lapla-
cian, it is usually difficult to compute them for a given Ω. In this course, we
shall compute the eigenvalues when Ω is a 2D-rectangle and a 2D-disk.
4.3.1 In Rectangle
Let the rectangle be Ω = {(x, y) ∈ R2 | 0 < x < a, 0 < y < b}. We wish to
solve the Dirichlet EVP in the rectangle Ω

−∆u(x, y) = λu(x, y) (x, y) ∈ Ω
u(x, y) = 0 (x, y) ∈ ∂Ω.
The boundary condition amounts to saying
u(x, 0) = u(a, y) = u(x, b) = u(0, y) = 0.
We look for solutions of the form u(x, y) = v(x)w(y) (variable separated).

Substituting u in separated form in the equation, we get
−v ′′ (x)w(y) − v(x)w′′ (y) = λv(x)w(y).
Hence
v ′′ (x) w′′ (y)
− =λ+ .
v(x) w(y)
Since LHS is function of x and RHS is function y and are equal they must
be some constant, say µ. We need to solve the EVP’s
−v ′′ (x) = µv(x) and − w′′ (y) = (λ − µ)w(y)
under the boundary conditions v(0) = v(a) = 0 and w(0) = w(b) = 0.

As seen before, while solving for v, we have trivial solutions for µ ≤ 0. If
√ √
µ > 0, then v(x) = c1 cos( µx)+c2 sin( µx). Using the boundary condition
√
v(0) = 0, we get c1 = 0. Now using v(a) = 0, we have c2 sin µa = 0. Thus,
√
either c2 = 0 or sin µa = 0. We have non-trivial solution, if c2 6= 0,
√ √
then µa = kπ or µ = kπ/a, for k ∈ Z. For each k ∈ N, we have
vk (x) = sin(kπx/a) and µk = (kπ/a)2 . We solve for w for each µk . For each
k, l ∈ N, we have wkl (y) = sin(lπy/b) and λkl = (kπ/a)2 + (lπ/b)2 . For each
k, l ∈ N, we have
ukl (x, y) = sin(kπx/a) sin(lπy/b)
and λkl = (kπ/a)2 + (lπ/b)2 .
4.3.2 In Disk
Let the disk of radius a be Ω = {(x, y) ∈ R2 | x2 + y 2 < a2 }. We wish to
solve the Dirichlet EVP in the disk Ω
 −1 ∂ 1 ∂2u
 r ∂r r ∂u
∂r
− r2 ∂θ2 = λu(r, θ) (r, θ) ∈ Ω
u(θ) = u(θ + 2π) θ∈R

u(a, θ) = 0 θ ∈ R.
We look for solutions of the form u(r, θ) = v(r)w(θ) (variable separated).

Substituting u in separated form in the equation, we get

w d dv v
− r − 2 w′′ (θ) = λv(r)w(θ).
r dr dr r
Hence dividing by vw and multiplying by r2 , we get

r d dv 1
− r − w′′ (θ) = λr2 .
v dr dr w

r d dv −1 ′′
r + λr2 = w (θ) = µ.
v dr dr w
Solving for non-trivial w, using the periodicity of w, we get for µ0 = 0,
w0 (θ) = a20 and for each k ∈ N, µk = k 2 and
wk (θ) = ak cos kθ + bk sin kθ.
For each k ∈ N ∪ {0}, we have the equation,

d dv
r r + (λr2 − k 2 )v = 0.
dr dr
√
Introduce change of variable x = λr and x2 = λr2 . Then
d d
r =x .
dr dx
rewriting the equation in new variable y(x)) = v(r)

d dy(x)
x x + (x2 − k 2 )y(x) = 0.
dx dx
Note that this none other than the Bessel’s equation. We already know that
for each k ∈ N ∪ {0}, we have the Bessel’s function Jk as a solution to the
Bessel’s
√ equation. Recall
√ the boundary condition on v, v(a) = 0. Thus,
y( λa) = 0. Hence λa should be a zero of the Bessel’s function.√
For each k ∈ N∪{0}, let zkl be the l-th zero of Jk , l ∈ N. Hence λa = zkl
2
and so λkl = zkl /a2 and y(x) = Jk (x). Therefore, v(r) = Jk (zkl r/a). For each
k ∈ N ∪ {0} and l ∈ N, we have
ukl (r, θ) = Jk (zkl r/a) sin(kθ) or Jk (zkl r/a) cos(kθ)

2
and λkl = zkl /a2 .
Chapter 5
Heat Equation
5.1 Derivation
Let a homogeneous material occupy a region Ω ⊂ Rn with C 1 boundary. Let
k denote the thermal conductivity (dimensionless quantity) and c be the heat
capacity of the material. Let u(x, t) be a function plotting the temperature
of the material at x ∈ Ω in time t. The thermal energy stored at x ∈ Ω, at
time t, is cu(x, t). If v(x, t) denotes the velocity of (x, t), by Fourier law, the
thermal energy changes following the gradients of temperature, i.e.,
cu(x, t)v(x, t) = −k∇u.
The thermal energy is the quantity that is conserved (conservation law) and
satisfies the continuity equation (1.4.1). Thus, we have
k
ut − ∆u = 0.
c
If the material occupying the region Ω is non-homogeneous, anisotropic,
the temperature gradient may generate heat in preferred directions, which
themselves may depend on x ∈ Ω. Thus, the conductivity of such a material
at x ∈ Ω, at time t, is given by a n × n matrix K(x, t) = (kij (x, t)). Thus, in
this case, the heat equation becomes,

1
ut − div K∇u = 0.
c
The heat equation is an example of a second order equation in divergence
form. The heat equation gives the temperature distribution u(x, t) of the
107
CHAPTER 5. HEAT EQUATION 108
material with conductivity k and capacity c. In general, we may choose

k/c = 1, since, for any k and c, we may rescale our time scale t 7→ (k/c)t.
The Cauchy initial value problem (IVP) of the heat equation,

ut (x, t) − ∆x u(x, t) = 0 in Rn × (0, ∞)
(5.1.1)
u(x, 0) = g(x) in Rn ,
where g ∈ C(Ω). We end this section with a remark that under the transfor-
mation t 7→ −t, in contrast to the wave equation, the heat equation changes
to a background equation. This signifies that the heat equation describes
irreversible process, i.e., it is not possible to find the distribution of temper-
ature at an earlier time t < t0 , if the distribution is given at t0 .
5.2 Boundary Conditions

To make the heat equation ut − ∆u = f in Ω × (0, T ) well-posed, where Ω is
a bounded open subset of Rn , we choose to specify the boundary condition
u(x, 0) = g(x) on Ω × {t = 0} and one of the following conditions on ∂Ω ×
(0, T ):
(i) (Dirichlet condition) u(x, t) = h(x, t);
(ii) (Neumann condition) ∇x u(x, t) · ν(x) = h(x, t), where nu(x) is the unit
outward normal of (x, t) ∈ ∂Ω × (0, T );
(iii) (Robin condition) ∇x u(x, t) · ν + cu(x, t) = h(x, t) for any c > 0.
(iv) (Mixed condition) u(x, t) = h(x, t) on Γ1 and ∇x u(x, t) · ν = h(x, t) on

Γ2 , where Γ1 ∪ Γ2 = ∂Ω × (0, T ) and Γ1 ∩ Γ2 = ∅.
5.3 Heat Flow on a Bar

The equation governing the heat propagation in a bar of length L is

∂u 1 ∂ ∂u
= κ(x)
∂t ρ(x)σ(x) ∂x ∂x
where σ(x) is the specific heat at x, ρ(x) is density of bar at x and κ(x) is
the thermal conductivity of the bar at x. If the bar is homogeneous, i.e, its
properties are same at every point, then

∂u κ ∂ 2u
=
∂t ρσ ∂x2
with ρ, σ, κ being constants.
Let L be the length of a homogeneous rod insulated along sides insulated
along sides and its ends are kept at zero temperature. Then the temperature
u(x, t) at every point of the rod, 0 ≤ x ≤ L and time t ≥ 0 is given by the
equation
2
∂u 2∂ u
=c
∂t ∂x2
where c is a constant.
The temperature zero at the end points is given by the Dirichlet boundary
condition
u(0, t) = u(L, t) = 0.
Also, given is the initial temperature of the rod at time t = 0, u(x, 0) = g(x),
where g is given (or known) such that g(0) = g(L) = 0. Given g : [0, L] → R
such that g(0) = g(L) = 0, we look for all the solutions of the Dirichlet
problem

 ut (x, t) − c2 uxx (x, t) = 0 in (0, L) × (0, ∞)
u(0, t) = u(L, t) = 0 in (0, ∞)

u(x, 0) = g(x) on [0, L].
We look for u(x, t) = v(x)w(t) (variable separated). Substituting u in sepa-
rated form in the equation, we get
v(x)w′ (t) = c2 v ′′ (x)w(t)
w′ (t) v ′′ (x)
= .
c2 w(t) v(x)
Since LHS is function of t and RHS is function x and are equal they must be
some constant, say λ. Thus,
w′ (t) v ′′ (x)
= = λ.
c2 w(t) v(x)
Thus we need to solve two ODE to get v and w,
w′ (t) = λc2 w(t)
and
v ′′ (x) = λv(x).
But we already know how to solve the eigenvalue problem involving v. For
each k ∈ N, we have the pair (λk , vk ) as solutions
to the EVP involving v,
2 2 kπx
where λk = −(kπ) /L and vk (x) = sin L some constants bk . For each
k ∈ N, we solve for wk to get
ln wk (t) = λk c2 t + ln α
2
where α is integration constant. Thus, wk (t) = αe−(kcπ/L) t . Hence,

kπx −(kcπ/L)2 t
uk (x, t) = vk (x)wk (t) = βk sin e ,
L
for some constants βk , is a solution to the heat equation. By superposition
principle, the general solution is
X∞ X∞
kπx −(kcπ/L)2 t
u(x, t) = uk (x, t) = βk sin e .
k=1 k=1
L
We now use the initial temperature of the rod, given as g : [0, L] → R to

find the particular solution of the heat equation. We are given u(x, 0) = g(x).
Thus,
X ∞
kπx
g(x) = u(x, 0) = βk sin
k=1
L
Since g(0) = g(L) = 0, we know that g admits a Fourier Sine expansion and
hence its coefficients βk are given as
Z
2 L kπx
βk = g(x) sin .
L 0 L
5.4 On a Circular Wire

We intend solve the heat equation in a circle (circular wire) of radius one
which is insulated along its sides. Then the temperature u(θ, t) at every
point of the circle, θ ∈ R and time t ≥ 0 is given by the equation
∂u ∂ 2u
= c2 2
∂t ∂θ
where c is a constant. We note that now u(θ, t) is 2π-periodic in the variable

θ. Thus,
u(θ + 2π, t) = u(θ, t) ∀θ ∈ R, t ≥ 0.
Let the initial temperature of the wire at time t = 0, be u(θ, 0) = g(θ), where
g is a given 2π-periodic function. Given a 2π-periodic function g : R → R,
we look for all solutions of

 ut (θ, t) − c2 uθθ (θ, t) = 0 in R × (0, ∞)
u(θ + 2π, t) = u(θ, t) in R × (0, ∞)

u(θ, 0) = g(θ) on R × {t = 0}.
We look for u(θ, t) = v(θ)w(t) with variables separated. Substituting for u
in the equation, we get
w′ (t) v ′′ (θ)
= = λ.
c2 w(t) v(θ)
For each k ∈ N ∪ {0}, the pair (λk , vk ) is a solution to the EVP where
λk = −k 2 and
vk (θ) = ak cos(kθ) + bk sin(kθ).
2
For each k ∈ N ∪ {0}, we get wk (t) = αe−(kc) t . For k = 0
u0 (θ, t) = a0 /2 (To maintain consistency with Fourier series)
and for each k ∈ N, we have
2 c2 t
uk (θ, t) = [ak cos(kθ) + bk sin(kθ)] e−k
Therefore, the general solution is
∞
a0 X 2 2
u(θ, t) = + [ak cos(kθ) + bk sin(kθ)] e−k c t .
2 k=1
We now use the initial temperature on the circle to find the particular solu-
tion. We are given u(θ, 0) = g(θ). Thus,
∞
a0 X
g(θ) = u(θ, 0) = + [ak cos(kθ) + bk sin(kθ)]
2 k=1
Since g is 2π-periodic it admits a Fourier series expansion and hence

Z
1 π
ak = g(θ) cos(kθ) dθ,
π −π
Z π
1
bk = g(θ) sin(kθ) dθ.
π −π
Note that as t → ∞ the temperature of the wire approaches a constant a0 /2.
Exercise 20. Solve the heat equation for 2D Rectangle and 2D Disk
5.5 Inhomogeneous Equation

In this section we solve the inhomogeneous heat equation, using Duhamel’s
principle. The Duhamel’s principle states that one can obtain a solution of
the inhomogeneous IVP for heat from its homogeneous IVP.
For a given f , let u(x, t) be the solution of the inhomogeneous heat equa-
tion,

 ut (x, t) − c2 ∆u(x, t) = f (x, t) in Ω × (0, ∞)
u(x, t) = 0 in ∂Ω × (0, ∞)

u(x, 0) = 0 in Ω.
As a first step, for each s ∈ (0, ∞), consider w(x, t; s) as the solution of the
homogeneous problem (auxiliary)
 s
 wt (x, t) − c2 ∆ws (x, t) = 0 in Ω × (s, ∞)
ws (x, t) = 0 in ∂Ω × (s, ∞)

ws (x, s) = f (x, s) on Ω × {s}.
Since t ∈ (s, ∞), introducing a change of variable r = t−s, we have ws (x, t) =

w(x, t − s) which solves

 wt (x, r) − c2 ∆w(x, r) = 0 in Ω × (0, ∞)
w(x, r) = 0 in ∂Ω × (0, ∞)

w(x, 0) = f (x, s) on Ω.
Duhamel’s principle states that

Z t Z t
s
u(x, t) = w (x, t) ds = w(x, t − s) ds
0 0
and u solves the inhomogenous heat equation. Suppose w is C 2 , we get

Z
∂ t
ut (x, t) = w(x, t − s) ds
∂t 0
Z t
d(t)
= wt (x, t − s) ds + w(x, t − t)
0 dt
d(0)
− w(x, t − 0)
Z dt
t
= wt (x, t − s) ds + w(x, 0)
0
Z t
= wt (x, t − s) ds + f (x, t).
0
Similarly, Z t
∆u(x, t) = ∆w(x, t − s) ds.
0
Thus,
Z t
2

ut − c ∆u = f (x, t) + wt (x, t − s) − c2 ∆w(x, t − s) ds
0
= f (x, t).
5.6 Steady State Equation

Consider the Cauchy problem with inhomogeneous Dirichlet boundary con-
ditions 
 ut (x, t) − c2 ∆u(x, t) =0 in Ω × (0, ∞)
u(x, t) = = φ(x, t) in ∂Ω × [0, ∞)

u(x, 0) = g(x) on Ω
such that, for all x ∈ ∂Ω,
g(x) = φ(x, 0), g ′′ (x) = φ′′ (x, 0).
The steady-state solution of the heat equation is defined as
v(x) = lim t → ∞u(x, t).
Note that v satisfies the equation ∆v = 0, since vt = 0. Further v satisfies
the boundary condition on ∂Ω as
v(x) = lim t → ∞φ(x, t).
5.7 Fundamental Solution of Heat Equation

We shall now derive the fundamental solution of the heat equation (5.1.1).
Taking Fourier transform both sides of the equation, we get
ut\ − ∆u = fˆ
c t) = fˆ(ξ, t)
ubt (ξ, t) − ∆u(ξ,
X n
ût (ξ, t) − i2 ξj2 û(ξ, t) = fˆ(ξ, t)
j=1
ût (ξ, t) + |ξ|2 û(ξ, t) = fˆ(ξ, t).

The solution of the above ODE is given by (cf. Appendix G)
Z t
−|ξ|2 t 2
û(ξ, t) = ĝ(ξ)e + e−|ξ| (t−s) fˆ(ξ, s) ds
0
where we have used the initial condition û(ξ, 0) = ĝ(ξ) of the ODE. Therefore,
by inverse Fourier formula,
Z Z t
−n/2 −|ξ|2 t iξ·x iξ·x−|ξ|2 (t−s) ˆ
u(x, t) = (2π) ĝ(ξ)e e + e f (ξ, s) ds dξ
Rn 0
Z Z
−n −iξ·y 2
= (2π) g(y)e dy e−|ξ| t eiξ·x dξ
Rn Rn
Z Z t Z
−n −iξ·y 2
+(2π) f (y, s)e dy eiξ·x−|ξ| (t−s) ds dξ
n n
Z R 0 Z R
−n iξ·(x−y)−|ξ|2 t
= (2π) g(y) e dξ dy
Rn Rn
Z Z t Z
−n iξ·(x−y)−|ξ|2 (t−s)
+(2π) f (y, s) e dξ ds dy
Rn 0 Rn
Z Z Z t
= g(y)K(x, y, t) dy + f (y, s)K(x, y, t − s) ds dy
Rn Rn 0
where Z
−n 2
K(x, y, t) = (2π) eiξ·(x−y)−|ξ| t dξ.
Rn
Note that

2
√
(x − y) √ (x − y) |x − y|2
iξ · (x − y) − |ξ| t = − ξ t − i √ · ξ t−i √ −
2 t 2 t 4t
√ √
and, set η = ξ t − i (x−y)
√
2 t
. Therefore, dη = tdξ. Using this substituion
in K and simplifying, we get
2 /4t
K(x, y, t) = (4πt)−n/2 e−|x−y|
called the heat kernel or the fundamental solution of heat equation. The
function K can be motivated in another way. Note that if u(x, t) is a solu-
tion of the heat equation, then (u ◦ Tλ )(x, t) is also a solution of the heat
equation, where Tλ (x, t) = (λx, λ2 t) is a linear transformation for any λ 6= 0.
This scaling or dilation is called the parabolic scaling. Thus, we look for a
solution u(x, t) = v(t)w(r2 /t), where r = |x|. Substituting this separation
2
of variable in the heat equation, we derive v(t) = t−n/2 and w(t) = e−r /4t .
This motivates us to define the fundamental solution as
(
1 n/2 −r 2 /4t
− 4πt e x ∈ Rn , t > 0
K(x, t) :=
0 x ∈ Rn , t < 0.
Chapter 6
The Laplacian
A general second order linear elliptic equation is of the form

Xn n
X
aij (x)uxi xj + bi (x)uxi + c(x)u(x) = f (x)
i,j=1 i=1
where A(x) = aij (x) is real, symmetric and positive definite. If A is a

constant matrix then with a suitable transformation one can rewrite
Xn
aij (x)uxi xj = ∆v
i,j=1
where v(x) := u(T x). We introduced P (cf.∂ 2 Chapter 1) Laplacian to be the

trace of the Hessain matrix, ∆ := ni=1 ∂x 2 . The Laplace operator usually
i
appears in physical models associated with dissipative effects (except wave
equation). The importance of Laplace operator can be realised by its ap-
pearance in various physical models. For instance, the heat equation
∂
− ∆,
∂t
the wave equation
∂2
− ∆,
∂t2
or the Schrödinger’s equation
∂
i + ∆.
∂t
The Laplacian is a linear operator, i.e., ∆(u+v) = ∆u+∆v and ∆(λu) =
λ∆u for any constant λ ∈ R.
117
CHAPTER 6. THE LAPLACIAN 118
6.1 Properties of Laplacian

The Laplace operator commutes with the translation and rotation opera-
tor. For any a ∈ Rn , we define the translation map Ta : C(Ω) → C(Ω) as
(Ta u)(x) = u(x + a). The invariance of Laplacian under translation means
that ∆ ◦ Ta = Ta ◦ ∆. For any u ∈ C 2 (Ω), (Ta u)xi (x) = uxi (x + a) and
(Ta u)xi xi (x) = uxi xi (x + a). Thus, ∆(Ta u)(x) = ∆u(x + a).
For any orthogonal n × n matrix O (O−1 = Ot ), we define R : C(Ω) →
C(Ω) as Ru(x) = u(Ox). The invariance of Laplacian Pn under rotation means
that ∆ ◦ R = R ◦ ∆. Let y = Ox. Then, yj = i=1 Oji xi and, by chain rule,
n
X n
∂yj X
(Ru)xi = uyj = uyj Oji .
j=1
∂x i j=1
Therefore, ∇x Ru = Ot ∇y u and
(∆ ◦ R)u(x) = ∇x Ru · ∇x (Ru) = Ot ∇y u · Ot ∇y u = OOt ∇y u · ∇y u = ∆y u.
But ∆y u = (∆u)(Ox) = (R ◦ ∆)u(x).
A radial function is constant on every sphere about the origin. Since
Laplacian commutes with rotations, it should map the class of all radial
functions to itself.
In cartesian coordiantes, the n-dimensional Laplacian is given as
Xn
∂2
∆ := 2
.
i=1
∂x i
In polar coordinates (2 dimensions), the Laplacian is given as

1 ∂ ∂ 1 ∂2
∆ := r + 2 2
r ∂r ∂r r ∂θ
where r is the magnitude component (0 ≤ r < ∞) and θ is the direction
component (0 ≤ θ < 2π). The direction component is also called the azimuth
angle or polar angle.
Exercise 21. Show that the two dimensional Laplacian has the representation

1 ∂ ∂ 1 ∂2
∆ := r + 2 2
r ∂r ∂r r ∂θ
in polar coordinates.
Proof. Using the fact that x = r cos θ and y = r sin θ, we have

∂x ∂y ∂u ∂u ∂u
= cos θ, = sin θ and = cos θ + sin θ .
∂r ∂r ∂r ∂x ∂y
Also,
∂ 2u 2
2 ∂ u
2
2 ∂ u ∂ 2u
= cos θ + sin θ + 2 cos θ sin θ .
∂r2 ∂x2 ∂y 2 ∂x∂y
Similarly,
∂x ∂y ∂u ∂u ∂u
= −r sin θ, = r cos θ and = r cos θ − r sin θ .
∂θ ∂θ ∂θ ∂y ∂x
Also,
1 ∂ 2u 2
2 ∂ u
2
2 ∂ u ∂ 2u 1 ∂u
= sin θ + cos θ − 2 cos θ sin θ − .
r2 ∂θ2 ∂x2 ∂y 2 ∂x∂y r ∂r
Therefore,
∂ 2u 1 ∂ 2u ∂ 2u ∂ 2 u 1 ∂u
+ = + θ − .
∂r2 r2 ∂θ2 ∂x2 ∂y 2 r ∂r
and hence
∂ 2u 1 ∂ 2 u 1 ∂u
∆u = + + .
∂r2 r2 ∂θ2 r ∂r
In cylindrical coordinates (3 dimensions), the Laplacian is given as

1 ∂ ∂ 1 ∂2 ∂2
∆ := r + 2 2+ 2
r ∂r ∂r r ∂θ ∂z
where r ∈ [0, ∞), θ ∈ [0, 2π) and z ∈ R. In spherical coordinates (3 dimen-

sions), the Laplacian is given as

1 ∂ 2 ∂ 1 ∂ ∂ 1 ∂2
∆ := 2 r + 2 sin φ + 2 2
r ∂r ∂r r sin φ ∂φ ∂φ r sin φ ∂θ2
where r ∈ [0, ∞), φ ∈ [0, π] (zenith angle or inclination) and θ ∈ [0, 2π)
(azimuth angle).
d2
Note that in one dimension, n = 1, ∆ = dx 2.
Proposition 6.1.1. Let n ≥ 2 and u be a radial function, i.e., u(x) = v(r)

where x ∈ Rn and r = |x|, then
d2 v(r) (n − 1) dv(r)
∆u(x) = + .
dr2 r dr
Proof. Note that
p
∂r ∂|x| ∂( x21 + . . . + x2n )
= =
∂xi ∂xi ∂xi
1 2
= (x1 + . . . + x2n )−1/2 (2xi )
2
xi
= .
r
Thus,
Xn Xn
∂ ∂u(x) ∂ dv(r) xi
∆u(x) = =
i=1
∂xi ∂xi i=1
∂xi dr r
Xn
∂ 1 dv(r) n dv(r)
= xi +
i=1
∂xi r dr r dr
n
X x2 d dv(r) 1
i n dv(r)
= +
i=1
r dr dr r r dr
Xn
x2i 1 d2 v(r) 1 dv(r) n dv(r)
= 2
− 2 +
i=1
r r dr r dr r dr

r2 1 d2 v(r) 1 dv(r) n dv(r)
= 2
− 2 +
r r dr r dr r dr
2
d v(r) 1 dv(r) n dv(r)
= − +
dr2 r dr r dr
2
d v(r) (n − 1) dv(r)
= + .
dr2 r dr
Hence the result proved.
More generally, the Laplacian in Rn may be written in polar coordinates
as
∂2 n−1 ∂ 1
∆ := 2
+ + 2 ∆Sn−1
∂r r ∂r r
where ∆Sn−1 is a second order differential operator in angular variables only.

The above forms of Laplacian gives an insight into how the Laplacian treats
radial function and angular functions. The angular part of Laplacian is
called the Laplace-Beltrami operator acting on Sn−1 (unit sphere of Rn ) with
Riemannian metric induced by the standard Euclidean metric in Rn .
6.2 Ill-Posedness of Cauchy Problem

Recall that for a second order Cauchy problem we need to know both u
and its normal derivative on a data curve Γ contained in Ω. However, the
Cauchy problem for Laplacian (more generally for elliptic equations) is not
well-posed. In fact, the Cauchy problem for Laplace equation on a bounded
domain Ω is over-determined.
Example 6.1 (Hadamard). Consider the Cauchy problem for Laplace equa-
tion 
 uxx + uyy = 0
u(0, y) = cosk2ky

ux (0, y) = 0,
where k > 0 is an integer. It is easy to verify that there is a unique solution
cosh(kx) cos(ky)
uk (x, y) =
k2
of the Cauchy problem. Note that for any x0 > 0,
cosh(kx0 )
|uk (x0 , nπ/k)| = .
k2
Since, as k → ∞, nπ/k → 0 and |uk (x0 , nπ/k)| → ∞ the Cauchy problem is
not stable, and hence not well-posed.
Exercise 22. Show that the Cauchy problem for Laplace equation

 uxx + uyy = 0
u(x, 0) =0

uy (x, 0) = k −1 sin kx,
where k > 0, is not well-posed. (Hint: Compute explicit solution using sep-
aration of variable. Note that, as k → ∞, the Cauchy data tends to zero
uniformly, but the solution does not converge to zero for any y 6= 0. There-
fore, a small change from zero Cauchy data (with corresponding solution
being zero) may induce bigger change in the solution.)
This issue of ill-posedness of the Cauchy problem is very special to second

order elliptic equations. In general, any hyperbolic equation Cauchy problem
is well-posed, as long as the hyperbolicity is valid in the full neighbourhood
of the data curve.
Example 6.2. Consider the Cauchy problem for the second order hyperbolic
equation  2
 y uxx − yuyy + 12 uy = 0 y > 0
u(x, 0) = f (x)

uy (x, 0) = g(x).
The general solution to this problem can be computed as

2 3/2 2 3/2
u(x, y) = F x + y +G x− y .
3 3
On y = 0 u(x, 0) = F (x) + G(x) = f (x). Further,

1/2 ′ 2 3/2 1/2 ′ 2 3/2
uy (x, y) = y F x + y −y G x− y
3 3
and uy (x, 0) = 0. Thus, the Cauchy problem has no solution unless g(x) = 0.
If g ≡ 0 then the solution is

2 3/2 2 3/2 2 3/2
u(x, y) = F x + y −F x− y +f x− y
3 3 3
for arbitrary F ∈ C 2 . Therefore, when g ≡ 0 the solution is not unique.

The Cauchy problem is not well-posed because the equation is hyperbolic
(B 2 − AC = y 3 ) not in the full neighbourhood of the data curve {y = 0}.
6.3 Boundary Conditions

To make the Poisson equation −∆u = f in Ω well-posed, where Ω is a
bounded open subset of Rn , we choose to specify one of the following condi-
tions on the boundary, ∂Ω, of Ω:
(i) (Dirichlet condition) u = g;
(ii) (Neumann condition) ∇u·ν = g, where ν(x) is the unit outward normal
of x ∈ ∂Ω;
(iii) (Robin condition) ∇u · ν + cu = g for any c > 0.
(iv) (Mixed condition) u = g on Γ1 and ∇u·ν = h on Γ2 , where Γ1 ∪Γ2 = ∂Ω

and Γ1 ∩ Γ2 = ∅.
The Poisson equation with Neumann boundary condition comes with a

compatibility condition. Note that, by Guass divergence theorem (cf. Corol-
lary D.0.7), if u is a solution of the Neumann problem then u satisfies, for
every connected component ω of Ω,
Z Z
∆u = ∇u · ν (Using GDT)
ω ∂ω
Z Z
− f = g.
ω ∂ω
The second equality is called the compatibility condition. Thus, for a Neu-
mann problem the given data f, g must necessarily satisfy the compatibility
condition. Otherwise, the Neumann problem does not make any sense.
The aim of this chapter is solve

−∆u(x) = f (x) in Ω
one of the above inhomogeneous boudary condition on ∂Ω,
for any open subset Ω ⊂ Rn . Note that, by linearity of Laplacian, u = v + w

where v is a harmonic function1 that solves

∆v(x) = 0 in Ω
one of the above inhomogeneous boudary condition on ∂Ω,
and w solves the Poisson equation2

−∆w(x) = f (x) in Ω
one of the above homogeneous boudary condition on ∂Ω.
Therefore, we shall restrict our attention to solving only for v and w. We

begin our analysis with v called harmonic functions.
1
defined later
2
defined later
6.4 Harmonic Functions

The one dimensional Laplace equation is a ODE and is solvable with solu-
tions u(x) = ax + b for some constants a and b. But in higher dimensions
solving Laplace equation is not so simple. For instance, a two dimensional
Laplace equation
uxx + uyy = 0
has the trivial solution, u(x, y) = ax + by + c, all one degree polynomials of
two variables. In addition, xy, x2 − y 2 , x3 − 3xy 2 , 3x2 y − y 3 , ex sin y and
ex cos y are all solutions to the two variable Laplace equation. In Rn , it is
trivial to check that all polynomials up to degree one, i.e.
X
aα x α
|α|≤1
is a solution to ∆u = 0 in Rn . But we also have functions of higher degree

and functions not expressible in terms of elementaryQfunctions as solutions
to Laplace equation. For instance, note that u(x) = ni=1 xi is a solution to
∆u = 0 in Rn .
Definition 6.4.1. Let Ω be an open subset of Rn . A function u ∈ C 2 (Ω) is
said to be harmonic on Ω if ∆u(x) = 0 in Ω.
We already remarked that every scalar potential is a harmonic function.
Gauss was the first to deduce some important properties of harmonic func-
tions and thus laid the foundation for Potential theory and Harmonic Anal-
ysis.
Due to the linearity of ∆, sum of any finite number of harmonic func-
tions is harmonic and a scalar multiple of a harmonic function is harmonic.
Moreover, harmonic functions can be viewed as the null-space of the Laplace
operator, say from C 2 (Ω) to C(Ω), the space of continuous functions.
In two dimension, one can associate with a harmonic function u(x, y), a
conjugate harmonic function, v(x, y) which satisfy the first order system of
PDE called the Cauchy-Riemann equations,
ux = v y and uy = −vx .
Harmonic functions and holomorphic functions (differentiable complex func-

tions) are related in the sense that, for any pair (u, v), harmonic and its conju-
gate, gives a holomorphic function f (z) = u(x, y) + iv(x, y) where z = x + iy.
Conversely, for any holomorphic function f , its real part and imaginary part
are conjugate harmonic functions. This observation gives us more examples
of harmonic functions, for instance, since all complex polynomials f (z) = z m
are holomorphic we have (using the polar coordinates) u(r, θ) = rm cos mθ
and v(r, θ) = rm sin mθ are harmonic functions in R2 for all m ∈ N. Simi-
larly, since f (z) = log z = ln r + iθ is holomorphic in certain region, we have
u(r, θ) = ln r and v(r, θ) = θ are harmonic in R2 \ (0, 0) and R2 \ {θ = 0},
respectively.
Exercise 23. Show that there are infinitely many linearly independent har-
monic functions in the vector space C 2 (R2 ).
6.4.1 Spherical Harmonics

A polynomial of degree k in n-variables is of the form
X
Pk (x) := aα x α .
|α|≤k
A polynomial P is said to be homogeneous of degree k if P (λx) = λk P (x)

for any real λ 6= 0. Note that a homogeneous polynomial of degree k is of
the form
X
aα x α .
|α|=k

The number of possible n-tuples α such that |α| = k is given by n+k−1k
. Let
n
Hk (R ) denote the set of all homogeneous harmonic polynomial of degree k
in n variables. Note that Hk (Rn ) forms a vector space. Recall that the class
of harmonic functions, call it N (∆), is a (vector) subspace of C 2 (Rn ) and,
hence, Hk (Rn ) ⊂ N (∆) ⊂ C 2 (Rn ).
Two Dimensions
Consider a general homogeneous polynomial
k
X
Pk (x, y) := ai xi y k−i
i=0
of degree k in R2 (two variables). Note that Pk contains k + 1 coefficients3 .

Then
k
X k−2
X
i−2 k−i
∆Pk (x, y) = ai i(i − 1)x y + ai (k − i)(k − i − 1)xi y k−i−2
i=2 i=0
is a homogeneous polynomial of degree k − 2 and, hence, contains k − 1

coefficients. If Pk ∈ Hk (R2 ), i.e. ∆Pk (x, y) = 0, then all the k − 1 coefficients
should vanish. Thus, we have k − 1 equations relating the k + 1 coefficients
of Pk and, hence, Hk (R2 ) is of dimension two (since k + 1 − (k − 1) = 2).
Let us now find the basis of the two dimensional space Hk (R2 ). In polar
coordinates, Pk (r, θ) = rk Qk (θ) where
k
X
Qk (θ) = ai (cos θ)i (sin θ)k−i .
i=0
Note that Qk is the restriction of Pk to S1 and are called spherical harmonics.

If Pk ∈ Hk (R2 ) then, using the polar form of Laplacian, we get

rk−2 Q′′k (θ) + k 2 Qk (θ) = 0.
Therefore, for all r > 0, Qk (θ) is a solution to the ODE
Q′′k (θ) + k 2 Qk (θ) = 0.
Therefore, Qk (θ) = α cos kθ + β sin kθ and Pk (r, θ) = rk (α cos kθ + β sin kθ).

Thus, Pk is a linear combination rk cos kθ and rk sin kθ. In fact, if we identify
each vector (x, y) ∈ R2 with the complex number z = x + iy, then we have
shown that Re(z k ), Im(z k ) are the basis of Hk (R2 ). If we choose α1 and
β1 such that β = −α1 sin β1 and α = α1 cos β1 , then we can rewrite the
polynomial as
Pk (r, θ) = α1 rk cos(kθ + β1 ).
Thus, we immediately see that the zero set of Pk (r, θ) in R2 will be a family of
k straight lines passing through origin such that between any two consecutive
lines the angle is same.
3 k+1

k =k+1
Three Dimensions
X
Pk (x) := aα x α
|α|=k
(k+2)(k+1)
of degree k in R3 (three variables). Note that Pk contains k+2 k
= 2
coefficients. Then ∆Pk (x) is a homogeneous polynomial of degree k − 2 and,
hence, contains k(k−1)
2
coefficients. If Pk ∈ Hk (R3 ), i.e. ∆Pk (x) = 0, then all
the k(k−1)
2
coefficients should vanish. Thus, we have k(k−1) 2
equations relating
(k+2)(k+1)
the 2
coefficients of Pk and, hence, Hk (R3 ) is of dimension
(k + 2)(k + 1) − k(k − 1)
= 2k + 1.
2
The basis of the 2k +1 dimensional space Hk (R3 ) is given in terms of the Leg-
endre functions which we shall describe now. In spherical coordinates, x =
r sin φ cos θ, y = r sin φ sin θ and z = r cos φ. Thus, Pk (r, φ, θ) = rk R(φ)Q(θ)
where
X
R(φ)Q(θ) = aα (sin φ)α1 +α2 (cos φ)α3 (cos θ)α1 (sin θ)α2 .
|α|=k
The separated variable assumption above is not a issue because differential

operator is linear. Note that RQ is the restriction of Pk to S2 and are
called spherical harmonics. If Pk ∈ Hk (R2 ) then, using the spherical form of
Laplacian, we get
′′

k−2 2 2 R (φ) R′ (φ) Q′′ (θ)
r k(k + 1) sin φ + sin φ + sin φ cos φ + = 0.
R(φ) R(φ) Q(θ)
Therefore, for all r > 0, we have equality
R′′ (φ) R′ (φ) Q′′ (θ)
k(k + 1) sin2 φ + sin2 φ + sin φ cos φ =− .
R(φ) R(φ) Q(θ)
Since LHS is a function of φ and RHS is a function of θ they must be equal
to some constant λ. Then, we have to solve for the eigenvalue problem
−Q′′ (θ) = λQ(θ)

where Q is 2π-periodic. This has the solution, for all m ∈ N ∪ {0}, λ = m2

and Qm (θ) = αm cos mθ + βm sin mθ. For λ = m2 we solve for R(φ) in

′′ cos φ ′ m2
R (φ) + R (φ) = R(φ) − k(k + 1) φ ∈ (0, φ).
sin φ sin2 φ
dw
Set w = cos φ. Then dφ
= − sin φ.
dR d2 R dR
R′ (φ) = − sin φ and R′′ (φ) = sin2 φ 2 − cos φ
dw dw dw
In the new variable w, we get the Legendre equation

2 ′′ ′ m2
(1 − w )R (w) − 2wR (w) = − k(k + 1) R(w) w ∈ [−1, 1].
1 − w2
For each k ∈ N ∪ {0}, this has the Legendre polynomials, Rk,m (cos φ), as its
solutions. Therefore, in general,
Pk (r, φ, θ) = rk (α cos mθ + β sin mθ)Rk,m (cos φ).
However, we are interested only those Rk,m which gives a polynomial of degree
k in R3 . Thus, for m = 0, 1, . . . , k,
dk+m
Rk,m (w) = (1 − w2 )m/2 k+m
(1 − w2 )k .
dw
Note that, for each fixed k and all 1 ≤ m ≤ k, the collection
{Rk,0 (cos φ), cos mθRk,m (cos φ), sin mθRk,m (cos φ)} ⊂ Hk (R3 )
is 2k + 1 linearly independent homogeneous harmonic polynomials of degree

k and forms a basis. Thus, each Pk is a linear combination of these basis
elements.
The zero sets of Pk exhibit properties depending on m. For m = 0 the
harmonic polynomial Pk is a constant multiple of Rk,0 (cos φ). Since Rk,0 (w)
has k distinct zeros in [−1, 1] arranged symmetrically about w = 0, there are
k distince zeros of Rk,0 (cos φ) in (0, π) arranged symmetrically about π/2.
Thus on S2 , the unit sphere, the function Rk,0 (cos φ) vanishes on k circles
circumscribed in the latitudinal direction. For k odd the circle along equator
is also a zero set. The function Rk,0 (cos φ) and its constant multiples are
called zonal harmonics.
If 0 < m < k, then the spherical harmonics is of the form
dk+m
(α cos mθ + β sin mθ) sinm φ (1 − w2 )k .
dwk+m
If the first term is zero then tan mθ = −α/β. This corresponds to great
circle through the north pole and south pole of S2 and the angle between
the planes containing two consecutive great circle is π/m. The second term
vanishes on φ = 0 and φ = π corresponding to the north and south pole,
respectively. The third term vanishes on k − m latitude circle. Thus, we have
orthogonally intersecting family of circles which form the zero set which are
called tesseral harmonics.
If m = k then the spherical harmonics is of the form
(α cos kθ + β sin kθ) sink φ
and it vanishes for φ = 0, φ = π or tan kθ = −α/β. The first two cases

corresponds to the north and south pole, respectively, and the last case cor-
responds to great circles through the north pole and south pole of S2 and
the angle between the planes containing two consecutive great circle is π/k.
Thus, the great circles divide the S2 in to 2k sectors and are called sectorial
harmoics.
Higher Dimensions
X
Pk (x) := aα x α
|α|=k

of degree k in Rn (n variables). Note that Pk contains n+k−1 k
coefficients.
Then ∆Pk (x) is a homogeneous polynomial of degree k − 2 and, hence, con-
n+k−3 n
tains k−2 coefficients. If Pk ∈ Hk (R ), i.e. ∆Pk (x) = 0, then all the
n+k−3
n+k−3

k−2
coefficients should vanish. Thus, we have k−2
equations relating
n+k−1 n
k
coefficients of Pk and, hence, Hk (R ) is of dimension

n+k−1 n+k−3
ℓ := − .
k k−2
In polar form, Pk (r, θ) = rk Q(θ) where θ ∈ S n−1 and if Pk (r, θ) ∈ Hk (Rn )

then
∂ 2 Pk n − 1 ∂Pk 1
∆Pk = 2
+ + 2 ∆Sn−1 Pk = 0
∂r r ∂r r
where ∆Sn−1 is a second order differential operator in angular variables only
called the Laplace-Beltrami operator. Therefore, we have
rk−2 [∆Sn−1 Q(θ) + k(n + k − 2)Q(θ)] = 0
and for r > 0,

∆Sn−1 Q(θ) + k(n + k − 2)Q(θ) = 0.
6.4.2 Properties of Harmonic Functions

In this section we shall study properties of harmonic functions. We shall as-
sume the divergence theorems from multivariable calculus (cf. Appendix D).
Also, note that if u is a harmonic function on Ω then, by Gauss divergence
theorem (cf. Theorem D.0.6),
Z
∂u
dσ = 0.
∂Ω ∂ν
2π n/2
Definition 6.4.2. Let Ω be an open subset of Rn and wn = Γ(n/2)
(cf. Ap-
pendix E) be the surface area of the unit sphere S1 (0) of Rn .
(a) A function u ∈ C(Ω) is said to satisfy the first mean value property
(I-MVP) in Ω if
Z
1
u(x) = u(y) dσy for any Br (x) ⊂ Ω.
ωn rn−1 Sr (x)
(b) A function u ∈ C(Ω) is said to satisfy the second mean value property
(II-MVP) in Ω if
Z
n
u(x) = u(y) dy for any Br (x) ⊂ Ω.
ωn rn Br (x)
Exercise 24. Show that u satisfies the I-MVP iff

Z
1
u(x) = u(x + rz) dσz .
ωn S1 (0)
Similarly, u satisfies II-MVP iff
Z
n
u(x) = u(x + rz) dz.
ωn B1 (0)
Exercise 25. Show that the first MVP and second MVP are equivalent. That
is show that u satisfies (a) iff u satisfies (b).
Owing to the above exercise we shall, henceforth, refer to the I-MVP and
II-MVP as just mean value property (MVP).
We shall now prove a result on the smoothness of a function satisfying
MVP.
Theorem 6.4.3. If u ∈ C(Ω) satisfies the MVP in Ω, then u ∈ C ∞ (Ω).
Proof. We first consider uε := ρε ∗ u, the convolution of u with mollifiers, as
introduced in Theorem F.0.10. where
Ωε := {x ∈ Ω | dist(x, ∂Ω) > ε}.
We shall now show that u = uε for all ε > 0, due to the MVP of u and the
radial nature of ρ. Let x ∈ Ωε . Consider
Z
uε (x) = ρε (x − y)u(y) dy
ZΩ
= ρε (x − y)u(y) dy (Since supp(ρε ) is in Bε (x))
Bε (x)
Z ε Z
= ρε (r) u(y) dσy dr (cf. Theorem E.0.8)
0 Sr (x)
Z ε
= u(x)ωn ρε (r)rn−1 dr (Using MVP of u)
Z ε0 Z
= u(x) ρε (r) dσy dr
0 Sr (0)
Z
= u(x) ρε (y) dy = u(x).
Bε (0)
Thus uε (x) = u(x) for all x ∈ Ωε and for all ε > 0. Since uε ∈ C ∞ (Ωε ) for
all ε > 0 (cf. Theorem F.0.10), we have u ∈ C ∞ (Ωε ) for all ε > 0.
Theorem 6.4.4. Let u be a harmonic function on Ω. Then u satisfies the

MVP in Ω.
Proof. Let Br (x) ⊂ Ω be any ball with centre at x ∈ Ω and for some r > 0.
For the given harmonic function u, we set
Z
1
v(r) := u(y) dσy .
ωn rn−1 Sr (x)
Note that v is not defined at 0, since r > 0. We have from Exercise 24 that
Z
1
v(r) = u(x + rz) dσz .
ωn S1 (0)
Now, differentiating both sides w.r.t r, we get
Z
dv(r) 1
= ∇u(x + rz) · z dσz
dr ωn S1 (0)
Z
1 (y − x)
= n−1
∇u(y) · dσy
ωn r Sr (x) r
Since |x − y| = r, by setting ν := (y − x)/r as the unit vector, and applying
the Gauss divergence theorem along with the fact that u is harmonic, we get
Z Z
dv(r) 1 1
= ∇u(y) · ν dσy = ∆u(y) dy = 0.
dr ωn rn−1 Sr (x) ωn rn−1 Br (x)
Thus, v is a constant function of r > 0 and hence
v(r) = v(ε) ∀ε > 0.
Moreover, since v is continuous (constant function), we have
v(r) = lim v(ε)
ε→0
Z
1
= lim u(x + εz) dσz
ε→0 ωn S (0)
1
Z
1
= lim u(x + εz) dσz (u is continuous on S1 (0))
ωn S1 (0) ε→0
Z
1
= u(x) dσz
ωn S1 (0)
= u(x) (Since ωn is the surface area of S1 (0)).
Thus, u satisfies I-MVP and hence the II-MVP.
Corollary 6.4.5. If u is harmonic on Ω, then u ∈ C ∞ (Ω).
The above corollary is a easy consequence of Theorem 6.4.4 and Theo-

rem 6.4.3. We shall now prove that any function satisfying MVP is harmonic.
Theorem 6.4.6. If u ∈ C(Ω) satisfies the MVP in Ω, then u is harmonic

in Ω.
Proof. Since u satisfies MVP, by Theorem 6.4.3, u ∈ C ∞ (Ω). Thus, ∆u

makes sense. Now, suppose u is not harmonic in Ω, then there is a x ∈ Ω
such that ∆u(x) 6= 0. Without loss of generality, let’s say ∆u(x) > 0.
Moreover, since ∆u is continuous there is a s > 0 such that, for all y ∈ Bs (x),
∆u(y) > 0. As done previously, we set for r > 0,
Z
1
v(r) := u(y) dσy .
ωn rn−1 Sr (x)
Thus, v(r) = u(x) for all r > 0 and hence v is a constant function of r and
v ′ (s) = 0. But
Z
dv(s) 1
0= = ∆u(y) dy > 0
dr ωn rn−1 Bs (x)
is a contradiction. Therefore, u is harmonic in Ω.
Above results leads us to conclude that a function is harmonic iff it sat-

isfies the MVP.
Exercise 26. If um is a sequence of harmonic functions in Ω converging to u
uniformly on compact subsets of Ω, then show that u is harmonic in Ω.
Theorem 6.4.7 (Strong Maximum Principle). Let Ω be an open, connected

(domain) subset of Rn . Let u be harmonic in Ω and M := maxy∈Ω u(y).
Then
u(x) < M ∀x ∈ Ω
or u ≡ M is constant in Ω.
Proof. We define a subset S of Ω as follows,
S := {x ∈ Ω | u(x) = M }.
If S = ∅, we have u(x) < M for all x ∈ Ω. Suppose S 6= ∅. Then S is closed

subset of Ω, since u is continuous. Now, for any x ∈ S, by MVP
Z
n
u(x) = u(y) dy for every r such that Br (x) ⊂ Ω.
ωn rn Br (x)
Thus, we have Z
n
M = u(x) = u(y) dy ≤ M
ωn rn Br (x)
Hence equality will hold above only when u(y) = M for all y ∈ Br (x). Thus,
we have shown that for any x ∈ S, we have Br (x) ⊂ S. Therefore, S is open.
Since Ω is connected, the only open and closed subsets are ∅ or Ω. Since
S was assumed to be non-empty, we should have S = Ω. Thus, u ≡ M is
constant in Ω.
Corollary 6.4.8 (Weak maximum Principle). Let Ω be an open, bounded
subset of Rn . Let u ∈ C(Ω) be harmonic in Ω. Then
max u(y) = max u(y).
y∈Ω y∈∂Ω
Proof. Let M := maxy∈Ω u(y). If there is a x ∈ Ω such that u(x) = M , then

u ≡ M is constant on the connected component of Ω containing x. Thus,
u = M on the boundary of the connected component which is a part of
∂Ω.
Aliter. Since ∂Ω ⊂ Ω, we have max∂Ω u ≤ maxΩ u. It only remains to
prove the other equality. For the given harmonic function u and for a fixed
ε > 0, we set vε (x) = u(x) + ε|x|2 , for each x ∈ Ω. For each x ∈ Ω,
∆vε = ∆u + 2nε > 0. Recall that4 if a function v attains local maximum
at a point x ∈ Ω, then in each direction its second order partial derivative
vxi xi (x) ≤ 0, for all i = 1, 2, . . . , n. Therefore ∆v(x) ≤ 0. Thus, we argue
that vε does not attain (even a local) maximum in Ω. But vε has to have a
maximum in Ω, hence it should be attained at some point x⋆ ∈ ∂Ω, on the
boundary. For all x ∈ Ω,
u(x) ≤ vε (x) ≤ vε (x⋆ ) = u(x⋆ ) + ε|x⋆ |2 ≤ max u(x) + ε max |x|2 .
x∈∂Ω x∈∂Ω
The above inequality is true for all ε > 0. Thus, u(x) ≤ maxx∈∂Ω u(x), for all
x ∈ Ω. Therefore, maxΩ u ≤ maxx∈∂Ω u(x). and hence we have equality.
4
v ∈ C 2 (a, b) has a local maximum at x ∈ (a, b) then v ′ (x) = 0 and v ′′ (x) ≤ 0
Theorem 6.4.9 (Estimates on derivatives). If u is harmonic in Ω, then

Ck
|Dα u(x)| ≤ kuk1,Br (x) ∀Br (x) ⊂ Ω and each |α| = k
rn+k
n
where the constants C0 = ωn
and Ck = C0 (2n+1 nk)k for k = 1, 2, . . ..
Proof. We prove the result by induction on k. Let k = 0. Since u is harmonic,
by II-MVP we have, for any Br (x) ⊂ Ω,
Z
n

|u(x)| = u(y) dy
ωn rn Br (x)
Z
n
≤ |u(y)| dy
ωn rn Br (x)
n C0
= n
kuk1,Br (x) = n kuk1,Br (x) .
ωn r r
Now, let k = 1. Observe that if u is harmonic then by differentiating the
Laplace equation and using the equality of mixed derivatives, we have uxi :=
∂u
∂xi
is harmoic, for all i = 1, 2, . . . , n. Now, by the II-MVP of uxi , we have
Z
n2n

|uxi (x)| = u x i
(y) dy
ωn r Br/2 (x)
n
Z
n2n

= uνi dσy (by Gauss-Green theorem)
ωn r Sr/2 (x)
n
2n
≤ kuk∞,Sr/2 (x) .
r
Thus, it now remains to estimate kuk∞,Sr/2 (x) . Let z ∈ Sr/2 (x), then
Br/2 (z) ⊂ Br (x) ⊂ Ω.
But, using k = 0 result, we have

C 0 2n C 0 2n
|u(z)| ≤ kuk1,B (z) ≤ kuk1,Br (x) .
rn r/2
rn
C0 2n
Therefore, kuk∞,Sr/2 (x) ≤ rn
kuk1,Br (x) and using this in the estimate of uxi ,
we get
C0 n2n+1
|uxi (x)| ≤ kuk1,Br (x) .
rn+1
Hence
C1
|Dα u(x)| ≤ kuk1,Br (x) for |α| = 1.
rn+1
Let now k ≥ 2 and α be a multi-index such that |α| = k. We assume the
induction hypothesis that the estimate to be proved is true for k − 1. Note
βu
that Dα u = ∂D ∂xi
for some i ∈ {1, 2, . . . , n} and |β| = k − 1. Moreover,
if u is harmonic then by differentiating the Laplace equation and using the
βu
equality of mixed derivatives, we have ∂D ∂xi
is harmoic for i = 1, 2, . . . , n.
Thus, following an earlier argument, we have

β n Z β
∂D u(x) nk ∂D u(y)
|Dα u(x)| = = dy
∂xi ωn rn Br/k (x) ∂xi
Z
nk n β

= D uν i dσ y
ωn r Sr/k (x)
n
nk β
≤ kD uk∞,Sr/k (x) .
r
It now only remains to estimate kDβ uk∞,Sr/k (x) . Let z ∈ Sr/k (x), then
B(k−1)r/k (z) ⊂ Br (x) ⊂ Ω. But, using induction hypothesis for k − 1, we
have
Ck−1 k n+k−1 Ck−1 k n+k−1

|Dβ u(z)| ≤ kuk1,B (z) ≤ kuk1,Br (x) .
((k − 1)r)n+k−1 (k−1)r/k
((k − 1)r)n+k−1
Therefore, using the above estimate for Dα u, we get
Ck−1 nk n+k
|Dα u(x)| ≤ kuk1,Br (x)
(k − 1)n+k−1 rn+k
C0 2(n+1)(k−1) nk (k − 1)k−1 k n+k
= kuk1,Br (x)
(k − 1)n+k−1 rn+k
n
C0 (2n+1 nk)k k 1
= kuk1,Br (x)
rn+k k−1 2n+1
n
C0 (2n+1 nk)k k 1
= n+k
kuk1,Br (x)
r 2(k − 1) 2
n
Ck k 1
≤ n+k
kuk1,Br (x) since ≤ 1.
r 2(k − 1) 2
Hence
Ck
|Dα u(x)| ≤ kuk1,Br (x) for |α| = k, ∀k ≥ 2.
rn+k
Theorem 6.4.10 (Liouville’s Theorem). If u is bounded and harmonic on

Rn , then u is constant.
Proof. For any x ∈ Rn and r > 0, we have the estimate on the first derivative
as,
C1
|∇u(x)| ≤ kuk1,Br (x)
rn+1
2n+1 n
= kuk1,Br (x)
ωn rn+1
2n+1 n
≤ n+1
kuk∞,Rn ωn rn
ωn r
2n+1
= kuk∞,Rn → 0 as r → ∞.
r
Thus, ∇u ≡ 0 in Rn and hence u is constant.
Exercise 27. Show that if u is harmonic in Ω, then u is analytic in Ω. (Hint:

Use the estimates on derivatives with Stirling’s formula and Taylor expan-
sion).
We end our discussion on the properties of harmonic function with Har-
nack inequality. The Harnack inequality states that non-negative harmonic
functions cannot be very large or very small at any point without being so
everywhere in a compact set containing that point.
Theorem 6.4.11 (Harnack’s Inequality). Let u be harmonic in Ω and u ≥ 0

in Ω, then for each connected open subset ω ⊂⊂ Ω there is a constant C > 0
(depending only on ω) such that
sup u(x) ≤ C inf u(x).

x∈ω x∈ω
In particular,
1
u(y) ≤ u(x) ≤ Cu(y) ∀x, y ∈ ω.
C
Proof. Set r := 14 dist(ω, ∂Ω). Let x, y ∈ ω such that |x − y| < r. By II-MVP,

Z
n
u(x) = u(z) dz
ωn 2n rn B2r (x)
Z
n 1
≥ u(z) dz = u(y).
ωn 2n rn Br (y) 2n
Thus, 1/2n u(y) ≤ u(x). Interchanging the role of x and y, we get 1/2n u(x) ≤
u(y). Thus, 1/2n u(y) ≤ u(x) ≤ 2n u(y) for all x, y ∈ ω such that |x − y| ≤ r.
Now, let x, y ∈ ω. Since ω is compact and connected in Ω, we can pick
points x = x1 , x2 , . . . , xm = y such that ∪m
i=1 Bi ⊃ ω, where Bi := Br/2 (xi )
and are sorted such that Bi ∩ Bi+1 6= ∅, for i = 1, 2, . . . , m − 1. Hence, note
that |xi+1 − xi | ≤ r. Therefore,
1 1 1 1
u(x) = u(x1 ) ≥ n
u(x2 ) ≥ 2n u(x3 ) ≥ . . . ≥ (m−1)n u(xm ) = (m−1)n u(y).
2 2 2 2
1
Thus, C can be chosen to be 2(m−1)n
.
6.5 Existence and Uniqueness

A consequence of the maximum principle is the uniqueness of the harmonic
functions.
Theorem 6.5.1 (Uniqueness of Harmonic Functions). Let Ω be an open,
bounded subset of Rn . Let u1 , u2 ∈ C 2 (Ω) ∩ C(Ω) be harmonic in Ω such that
u1 = u2 on ∂Ω, then u1 = u2 in Ω.
Proof. Note that u1 −u2 is a harmonic function and hence, by weak maximum
principle, should attain its maximum on ∂Ω. But u1 − u2 = 0 on ∂Ω. Thus
u1 − u2 ≤ 0 in Ω. Now, repeat the argument for u2 − u1 , we get u2 − u1 ≤ 0
in Ω. Thus, we get u1 − u2 = 0 in Ω.
Let u ∈ C 2 (Ω) ∩ C(Ω) be a solution of the Dirichlet problem

∆u(x) = 0 x∈Ω
(6.5.1)
u(x) = g(x) x ∈ ∂Ω.
By the strong maximum principle (cf. Theorem 6.4.7), if Ω is connected and

g ≥ 0 and g(x) > 0 for some x ∈ ∂Ω then u(x) > 0 for all x ∈ Ω.
Theorem 6.5.2. Let Ω be an open bounded connected subset of Rn and

g ∈ C(∂Ω). Then the Dirichlet problem (6.5.1) has atmost one solution
u ∈ C 2 (Ω) ∩ C(Ω). Moreover, if u1 and u2 are solution to the Dirichlet
problem corresponding to g1 and g2 in C(∂Ω), respectively, then
(a) (Comparison) g1 ≥ g2 on ∂Ω and g1 (x0 ) > g2 (x0 ) for some x ∈ ∂Ω

implies that u1 > u2 in Ω.
(b) (Stability) |u1 (x) − u2 (x)| ≤ maxy∈∂Ω |g1 (y) − g2 (y)| for all x ∈ Ω.
Proof. The fact that there is atmost one solution to the Dirichlet problem
follows from the Theorem 6.5.1. Let w = u1 − u2 . Then w is harmonic.
(a) Note that w = g1 − g2 ≥ 0 on ∂Ω. Since g1 (x0 ) > g2 (x0 ) for some
x0 ∈ ∂Ω, then w(x) > 0 for all x ∈ ∂Ω. This proves the comparison
result.
(b) Again, by maximum principle, we have
±w(x) ≤ max |g1 (y) − g2 (y)|∀x ∈ Ω.

y∈∂Ω
This proves the stability result.
We remark that the uniqueness result is not true for unbounded domains.
Example 6.3. Consider the problem (6.5.1) with g ≡ 0 in the domain Ω =
{x ∈ Rn | |x| > 1}. Obviously, u = 0 is a solution. But we also have a
non-trivial solution (
ln |x| n=2
u(x) =
|x|2−n − 1 n ≥ 3.
Example 6.4. Consider the problem (6.5.1) with g ≡ 0 in the domain Ω =
{x ∈ Rn | xn > 0}. Obviously, u = 0 is a solution. But we also have a
non-trivial solution u(x) = xn .
We have shown above that if a solution exists for (6.5.1) then it is unique
(cf. Theorem 6.5.1). So the question that remains to be answered is on
the existence of solution of (6.5.1), for any given domain Ω. In the modern
theory, there are three different methods to address this question of existence,
viz., Perron’s Method, Layer Potential (Integral Equations) and L2 methods.
6.6 Perron’s Method

Definition 6.6.1. We say a function w ∈ C(Ω) is a barrier at x0 ∈ ∂Ω if
there is a neighbourhood U of x0 such that
1. w is superharmonic in Ω ∩ U
2. w > 0 in (Ω ∩ U ) \ {x0 } and w(x0 ) = 0.
Definition 6.6.2. Any point on ∂Ω is said to be regular (w.r.t Laplacian)

if there exists a barrier at that point.
A necessary and sufficient condition for the existence of solution to (6.5.1)

is given by the following result:
Theorem 6.6.3. The Dirichlet problem (6.5.1) is solvable for any arbitrary
bounded domain Ω and for any arbitrary g on ∂Ω iff all the points in ∂Ω are
regular.
Proof. One way is obvious. If (6.5.1) is solvable and x0 ∈ ∂Ω. Then, the
solution to
∆w = 0 in Ω
w = g on ∂Ω,
where g(x) = |x − x0 |, is a barrier function at x0 . Thus, any x0 ∈ ∂Ω
is a regular point. The converse is proved using the Perron’s method for
subharmonic functions.
Definition 6.6.4. A bounded domain Ω ⊂ Rn is said to satisfy the exterior

sphere condition if for every point x0 ∈ ∂Ω there is a ball B := BR (y) such
that B ∩ Ω = x0 .
Lemma 6.6.5. If Ω satisfies the exterior sphere condition then all boundary
points of Ω are regular.
Proof. For any x0 ∈ ∂Ω, we define the barrier function at x0 ∈ ∂Ω as

( 2−n 2−n
R − |x − y| for n ≥ 3
w(x) = |x−y|
ln R for n = 2.
Theorem 6.6.6. Any bounded domain with C 2 boundary satisfies the exte-
rior sphere condition.
Definition 6.6.7. A bounded domain Ω ⊂ Rn is said to satisfy the exterior
cone condition if for every point x0 ∈ ∂Ω there is a finite right circular cone
K with vertex at x0 such that K ∩ Ω = x0 .
Exercise 28. Any domain satisfying the exterior sphere condition also satisfies
the exterior cone condition.
Exercise 29. Every bounded Lipschitz domain satisfies the exterior cone con-
dition.
Lemma 6.6.8. If Ω satisfies the exterior cone condition then all boundary
points of Ω are regular.
6.6.1 Non-existence of Solutions

In 1912, Lebesgue gave an example of a domain on which the classical Dirich-
let problem is not solvable. Let
Ω := {(x, y, z) ∈ R3 | r2 + z 2 < 1; r > e−1/2z for z > 0}.
Note that Ω is the unit ball in R3 with a sharp inward cusp, called Lebesgue
spine, at (0, 0, 0). The origin is a not regular point of Ω.
However, there do exist domains with inward cusps for which the classical
problem is solvable, for instance, consider
Ω := {(x, y, z) ∈ R3 | r2 + z 2 < 1; r > z 2k for z > 0},
for any positive integer k. The proof of this fact involves the theory of
capacities.
6.6.2 Characterizing regular points

The Wiener’s criterion gives the necessary and sufficient condition for the
regularity of the boundary points. For n ≥ 3 and a fixed λ ∈ (0, 1), the
Wiener’s criterion states that a point x0 ∈ ∂Ω is regular iff the series
∞
X Ci
λ i(n−2)
i=0
/ Ω | |x − x0 | ≤ λi }.
diverges, where Ci := cap2 {x ∈
6.7 Ω with Simple Geometry

The method of separation of variables was introduced by d’Alembert (1747)
and Euler (1748) for the wave equation. This technique was also employed by
Laplace (1782) and Legendre (1782) while studying the Laplace equation and
also by Fourier while studying the heat equation. The motivation behind the
“separation of variable” technique will be highlighted while studying wave
equation.
Theorem 6.7.1 (2D Rectangle). Let Ω = {(x, y) ∈ R2 | 0 < x < a and 0 <
y < b} be a rectangle in R2 . Let g : ∂Ω → R which vanishes on three sides
of the rectangle, i.e., g(0, y) = g(x, 0) = g(a, y) = 0 and g(x, b) = h(x) where
h is a continuous function h(0) = h(a) = 0. Then there is a unique solution
to (6.5.1) on this rectangle with given boundary value g.
Proof. We begin by looking for solution u(x, y) whose variables are separated,
i.e., u(x, y) = v(x)w(y). Substituting this form of u in the Laplace equation,
we get
v ′′ (x)w(y) + v(x)w′′ (y) = 0.
Hence
v ′′ (x) w′′ (y)
=− .
v(x) w(y)
Since LHS is function of x and RHS is function y, they must equal a constant,
say λ. Thus,
v ′′ (x) w′′ (y)
=− = λ.
v(x) w(y)
Using the boundary condition on u, u(0, y) = g(0, y) = g(a, y) = u(a, y) =
0, we get v(0)w(y) = v(a)w(y) = 0. If w ≡ 0, then u ≡ 0 which is not a
solution to (6.5.1). Hence, w 6≡ 0 and v(0) = v(a) = 0. Thus, we need to
solve, ′′
v (x) = λv(x), x ∈ (0, a)
v(0) = v(a) = 0,
the eigen value problem for the second order differential operator. Note that
the λ can be either zero, positive or negative.
If λ = 0, then v ′′ = 0 and the general solution is v(x) = αx + β, for some
constants α and β. Since v(0) = 0, we get β = 0, and v(a) = 0 and a 6= 0
implies that α = 0. Thus, v ≡ 0 and hence u ≡ 0. But, this can not be a
solution to (6.5.1).
√ √
λx
If λ > 0, then v(x) = αe + βe− λx
. Equivalently,
√ √
v(x) = c1 cosh( λx) + c2 sinh( λx)
such that α = (c1 + c2 )/2 and β = (c1 − c2 )/2. Using the boundary condition
v(0) = 0, we get c1 = 0 and hence
√
v(x) = c2 sinh( λx).
√
Now using v(a) = 0, we have c2 sinh λa = 0. Thus, c2 = 0 and v(x) = 0.
We have seen this cannot √be a solution.
If λ < 0, then set ω = −λ. We need to solve
′′
v (x) + ω 2 v(x) = 0 x ∈ (0, a)
(6.7.1)
v(0) = v(a) = 0.
v(x) = α cos(ωx) + β sin(ωx).
Using the boundary condition v(0) = 0, we get α = 0 and hence v(x) =

β sin(ωx). Now using v(a) = 0, we have β sin ωa = 0. Thus, either β = 0
or sin ωa = 0. But β = 0 does not yield a solution. Hence ωa = kπ or
ω = kπ/a, for all non-zero k ∈ Z. Hence, for each k ∈ N, there is a solution
(vk , λk ) for (6.7.1), with

kπx
vk (x) = βk sin ,
a
for some constant βk and λk = −(kπ/a)2 . We now solve w corresponding to

each λk . For each k ∈ N, we solve for wk in the ODE
2
wk′′ (y) = kπ
a
wk (y), y ∈ (0, b)
w(0) = 0.
Thus, wk (y) = ck sinh(kπy/a). Therefore, for each k ∈ N,

kπx kπy
uk = δk sin sinh
a a
is a solution to (6.5.1). The general solution is of the form (principle of

superposition) (convergence?)
∞
X
kπx kπy
u(x, y) = δk sin sinh .
k=1
a a
The constant δk are obtained by using the boundary condition u(x, b) = h(x)
which yields
∞
X
kπb kπx
h(x) = u(x, b) = δk sinh sin .
k=1
a a
Since h(0) = h(a) = 0, we know that h admits a Fourier Sine series. Thus
δk sinh kπb
a
is the k-th Fourier sine coefficient of h, i.e.,
−1 Z a
kπb 2 kπx
δk = sinh h(x) sin .
a a 0 a
Theorem 6.7.2 (2D Disk). Let Ω = {(x, y) ∈ R2 | x2 + y 2 < R2 } be the disk

of radius R in R2 . Let g : ∂Ω → R is a continuous function. Then there is
a unique solution to (6.5.1) on the unit disk with given boundary value g.
Proof. Given the nature of the domain, we shall use the Laplace operator in
polar coordinates,
1 ∂ ∂ 1 ∂2
∆ := r + 2 2
r ∂r ∂r r ∂θ
where r is the magnitude component and θ is the direction component. Then
∂Ω is the circle of radius one. Then, solving for u(x, y) in the Dirichlet
problem is to equivalent to finding U (r, θ) : Ω → R such that
 1∂ 1 ∂2U
 r ∂r r ∂U
∂r
+ r2 ∂θ2 = 0 in Ω
U (r, θ + 2π) = U (r, θ) in Ω (6.7.2)

U (R, θ) = G(θ) on ∂Ω
where U (r, θ) = u(r cos θ, r sin θ), G : [0, 2π) → R is G(θ) = g(cos θ, sin θ).
Note that both U and G are 2π periodic w.r.t θ. We will look for solution
U (r, θ) whose variables can be separated, i.e., U (r, θ) = v(r)w(θ) with both
v and w non-zero. Substituting it in the polar form of Laplacian, we get

w d dv v d2 w
r + 2 2 =0
r dr dr r dθ
and hence
−r d dv 1 d2 w
r = .
v dr dr w dθ2
Since LHS is a function of r and RHS is a function of θ, they must equal a
constant, say λ. We need to solve the eigen value problem,
′′
w (θ) − λw(θ) = 0 θ∈R
w(θ + 2π) = w(θ) ∀θ.
Note that the λ can be either zero, positive or negative. If λ = 0, then

w′′ = 0 and the general solution is w(θ) = αθ + β, for some constants α and
β. Using the periodicity of w,
αθ + β = w(θ) = w(θ + 2π) = αθ + 2απ + β
implies that α = 0. Thus, the pair λ = 0 and w(θ) = β is a solution. If

λ > 0, then √ √
w(θ) = αe λθ + βe− λθ .
If either of α and β is non-zero, then w(θ) → ±∞ as θ → ∞, which contra-
dicts the periodicity of w. Thus,√α = β = 0 and w ≡ 0, which cannot be a
solution. If λ < 0, then set ω = −λ and the equation becomes
′′
w (θ) + ω 2 w(θ) = 0 θ∈R
w(θ + 2π) = w(θ) ∀θ
Its general solution is
w(θ) = α cos(ωθ) + β sin(ωθ).
Using the periodicity of w, we get ω = k where k is an integer. For each

k ∈ N, we have the solution (wk , λk ) where
λk = −k 2 and wk (θ) = αk cos(kθ) + βk sin(kθ).

For the λk ’s, we solve for vk , for each k = 0, 1, 2, . . .,

d dvk
r r = k 2 vk .
dr dr
For k = 0, we get v0 (r) = α ln r + β. But ln r blows up as r → 0, but any
solution U and, hence v, on the closed unit disk (compact subset) has to be
bounded. Thus, we must have the α = 0. Hence v0 ≡ β. For k ∈ N, we need
to solve for vk in
d dvk
r r = k 2 vk .
dr dr
ds d d ds
Use the change of variable r = es . Then es dr = 1 and dr = ds dr
= e1s ds
d
.
d d s ks −ks k −k −k
Hence r dr = ds . vk (e ) = αe + βe . vk (r) = αr + βr . Since r blows
up as r → 0, we must have β = 0. Thus, vk = αrk . Therefore, for each
k = 0, 1, 2, . . .,
Uk (r, θ) = ak rk cos(kθ) + bk rk sin(kθ).
∞
a0 X
U (r, θ) = + ak rk cos(kθ) + bk rk sin(kθ) .
2 k=1
To find the constants, we must use U (R, θ) = G(θ). If G ∈ C 1 [0, 2π], then G
admits Fourier series expansion. Therefore,
∞
a0 X k
G(θ) = + R ak cos(kθ) + Rk bk sin(kθ)
2 k=1
where Z π
1
ak = k G(θ) cos(kθ) dθ,
R π −π
Z π
1
bk = k G(θ) sin(kθ) dθ.
R π −π
Using this in the formula for U and the uniform convergence of Fourier series,
we get
Z " #
1 X r k
∞
1 π
U (r, θ) = G(η) + (cos kη cos kθ + sin kη sin kθ) dη
π −π 2 k=1 R
Z " #
1 X r k
∞
1 π
= G(η) + cos k(η − θ) dη.
π −π 2 k=1 R
Using the relation

∞
" ∞
# " #
X r k X r k r i(η−θ)
e
cos k(η − θ) = Re ei(η−θ) = Re R
k=1
R k=1
R 1 − Rr ei(η−θ)
2
R − rR cos(η − θ)
= −1
R2
+ r2 − 2rR cos(η − θ)
rR cos(η − θ) − r2
=
R2 + r2 − 2rR cos(η − θ)
in U (r, θ) we get
Z π
R2 − r 2 G(η)
U (r, θ) = dη.
2π −π R2 + r2 − 2rR cos(η − θ)
Note that the formula derived above for U (r, θ) can be rewritten in Carte-
sian coordinates and will have the form
Z
R2 − |x|2 g(y)
u(x) = 2
dy.
2πR SR (0) |x − y|
This can be easily seen, by setting y = R(x10 cos η +x20 sin η), we get dy = Rdη
and |x − y|2 = R2 + r2 − 2rR cos(η − θ). This is called the Poisson formula.
More generally, the unique solution to the Dirichlet problem on a ball of
radius R centred at x0 in Rn is given by Poisson formula
Z
R2 − |x − x0 |2 g(y)
u(x) = n
dy.
ωn R SR (x0 ) |x − y|
We will derive this general form later (cf. (6.8.7)).

Theorem 6.7.3 (3D Sphere). Let Ω = {(x, y, z) ∈ R3 | x2 + y 2 + z 2 < 1} be
the unit sphere in R3 . Let g : ∂Ω → R is a continuous function. Then there
is a unique solution to (6.5.1) on the unit sphere with given boundary value
g.
Proof. Given the nature of domain, the Laplace operator in spherical coor-
dinates,

1 ∂ 2 ∂ 1 ∂ ∂ 1 ∂2
∆ := 2 r + 2 sin φ + 2 2 .
r ∂r ∂r r sin φ ∂φ ∂φ r sin φ ∂θ2
where r is the magnitude component, φ is the inclination (zenith or elevation)

in the vertical plane and θ is the azimuth angle (in the direction in horizontal
plane). Solving for u in (6.5.1) is equivalent to finding U (r, φ, θ) : Ω → R
such that

 1 ∂ 2 ∂U 1 ∂ ∂U
 r2 ∂r r ∂r
+ r 2 sin φ ∂φ
sin φ ∂φ
1 2
+ r2 sin2 φ ∂∂θU2 = 0 in Ω (6.7.3)


U (1, φ, θ) = G(φ, θ) on ∂Ω
where U (r, φ, θ) and G(φ, θ) are appropriate spherical coordinate function

corresponding to u and g. We will look for solution U (r, φ, θ) whose variables
can be separated, i.e., U (r, φ, θ) = v(r)w(φ)z(θ) with v, w and z non-zero.
Substituting it in the spherical form of Laplacian, we get

wz d 2 dv vz d dw vw d2 z
r + sin φ + =0
r2 dr dr r2 sin φ dφ dφ r2 sin2 φ dθ2
and hence

1 d 2 dv −1 d dw 1 d2 z
r = sin φ − .
v dr dr w sin φ dφ dφ z sin2 φ dθ2
Since LHS is a function of r and RHS is a function of (φ, θ), they must equal
a constant, say λ. If Azimuthal symmetry is present then z(θ) is constant
and hence dz
dθ
= 0. We need to solve for w,
sin φw′′ (φ) + cos φw′ (φ) + λ sin φw(φ) = 0, φ ∈ (0, π)

dx
Set x = cos φ. Then dφ
= − sin φ.
dw d2 w dw
w′ (φ) = − sin φ and w′′ (φ) = sin2 φ 2 − cos φ
dx dx dx
In the new variable x, we get the Legendre equation
(1 − x2 )w′′ (x) − 2xw′ (x) + λw(x) = 0 x ∈ [−1, 1].
We have already seen that this is a singular problem (while studying S-L
problems). For each k ∈ N ∪ {0}, we have the solution (wk , λk ) where
λk = k(k + 1) and wk (φ) = Pk (cos φ).

For the λk ’s, we solve for vk , for each k = 0, 1, 2, . . .,

d 2 dvk
r = k(k + 1)vk .
dr dr
For k = 0, we get v0 (r) = −α/r + β. But 1/r blows up as r → 0 and U must
be bounded in the closed sphere. Thus, we must have the α = 0. Hence
v0 ≡ β. For k ∈ N, we need to solve for vk in

d 2 dvk
r = k(k + 1)vk .
dr dr
ds d d ds
Use the change of variable r = es . Then es dr = 1 and dr = ds dr
= e1s ds
d
.
d d 2
Hence r dr = ds . Solving for m in the quadratic equation m + m = k(k + 1).
m1 = k and m2 = −k − 1. vk (es ) = αeks + βe(−k−1)s . vk (r) = αrk + βr−k−1 .
Since r−k−1 blows up as r → 0, we must have β = 0. Thus, vk = αrk .
Therefore, for each k = 0, 1, 2, . . .,
Uk (r, φ, θ) = ak rk Pk (cos φ).

∞
X
U (r, φ, θ) = ak rk Pk (cos φ).
k=0
Since we have azimuthal symmetry, G(φ, θ) = G(φ). To find the constants,

we use U (1, φ, θ) = G(φ), hence
∞
X
G(φ) = ak Pk (cos φ).
k=0
Using the orthogonality of Pk , we have

Z
2k + 1 π
ak = G(φ)Pk (cos φ) sin φ dφ.
2 0
Now that we have sufficient understanding of harmonic functions, solution

of homogeneous Laplace equation, with Dirichlet boundary conditions we
next attempt to solve the inhomogeneous Laplace equation (called Poisson
equation) with homogeneous boundary conditions.
6.8 Poisson Equation

We now wish to solve the Poisson equation, for any given f (under some
hypothesis) find u such that
−∆u = f in Rn . (6.8.1)
Recall that we have already introduced the notion convolution of functions

(cf. Appendix F) while discussing C ∞ properties of harmonic functions. We
also observed that the differential operator can be accumulated on either side
of the convolution operation. Suppose there is a function K with the property
that ∆K is the identity of the convolution operation, i.e., f ∗ ∆K = f , then
we know that u := f ∗ K is a solution of (6.8.1).
Definition 6.8.1. We shall say a function K to be the fundamental solution

of the Laplacian, ∆, if ∆K is the identity with respect to the convolution
operation.
We caution that the above definition is not mathematically precise be-

cause we made no mention on what the “function” K could be and its dif-
ferentiability, even its existence is under question. We shall just take it as a
informal definition.
We note the necessry condition for any K to be a fundamental solution.
Observe that K is such that f ∗ ∆K for all f in the given space of functions
in Rn . In particular, one can choose f ≡ 1. Thus, the necessary condition
for a fundamental solution is 1 ∗ ∆K = 1, i.e.,
Z
∆K(x) dx = 1.
Rn
Equivalently, the necessary condition for K is

Z
lim ∆K(x) dx = 1,
r→∞ Br (0)
which by Gauss divergence theorem (all informally) means

Z
lim ∇K(y) · ν(y) dσy = 1.
r→∞ Sr (0)
6.8.1 Fundamental Solution of Laplacian

The invariance of Laplacian under rotation motivates us to look for a ra-
dial fundamental solution. Recall how Laplacian treats radial functions (cf.
Proposition 6.1.1) and, consequently, we have
Corollary 6.8.2. The function u(x) = ax + b solves ∆u = 0 in R. For
n ≥ 2, if u is a radial function on Rn then ∆u = 0 on Rn \ {0} iff
(
a + b ln |x| if n = 2,
u(x) = b
a + 2−n |x|2−n if n ≥ 3
where a, b are some constants.

Proof. For radial functions u(x) = v(r) where r = |x|. Observe that ∆u(x) =
0 iff v ′′ (r) + (n−1)
r
v ′ (r) = 0. Now, integrating both sides w.r.t r, we get
v ′′ (r) (1 − n)
=
v ′ (r) r
′
ln v (r) = (1 − n) ln r + ln b
v ′ (r) = br(1−n)
Integration both sides, once again, yields

(
b ln r + a if n = 2
v(r) = b 2−n
2−n
r + a if n 6= 2.
The reason to choose the domain of the Laplacian as Rn \ {0} is because

the operator involves a ‘r’ in the denominator. However, for one dimensional
case we can let zero to be on the domain of Laplacian, since for n = 1, the
Laplace operator is unchanged. Thus, for n = 1, u(x) = a + bx is a harmonic
function in Rn .
Note that as r → 0, v(r) → ∞. Thus, u has a singularity at 0. In fact,
for any given vector x0 ∈ Rn , ∆u(x − x0 ) = 0 for all x ∈ Rn \ {x0 }. We shall
choose a, b such that for every sphere Sr (0) about the origin, we have
Z
v ′ (r) dσ = 1.
Sr (0)
Thus,
Z (
b
′ r
(2πr) for n = 2
1= v (r) dσ = 1−n n−1
Sr (0) br (r ωn ) for n ≥ 3.
This is possible only for the choice

(
1
2π
for n = 2
b= 1
ωn
for n ≥ 3.
The constant a can be chosen arbitrarly, but to keep things simple, we choose
a ≡ 0 for n ≥ 2. For convention sake, we shall add minus (“−”) sign (notice
the minus sign in (6.8.1)).
Definition 6.8.3. For any fixed x0 ∈ Rn We say K(x0 , x), defined as
(
1
− 2π ln |x − x0 | (n = 2)
K(x0 , x) := |x−x0 |2−n
ωn (n−2)
(n ≥ 3),
is the fundamental solution of ∆ at any given x0 ∈ Rn .

We end this section by emphasising that the notion of fundamental so-
lution has a precise definition in terms of the Dirac measure. The Dirac
measure, at a point x ∈ Rn , is defined as,
(
1 if x ∈ E
δx (E) =
0 if x ∈
/E
for all measurable subsets E of the measure space Rn . The Dirac measure
has the property that Z
dδx = 1
E
if x ∈ E and zero if x ∈
/ E. Also, for any integrable function f ,
Z
f (y) dδx = f (x).
Rn
In this new set-up a fundamental solution K(x0 , ·) can be defined as the

solution corresponding to δx0 , i.e.,
−∆K(x0 , x) = δx0 in Rn .
Note that the above equation, as such, makes no sense because the RHS is a
set-function taking subsets of Rn as arguments, whereas K is a function on
Rn . To give meaning to above equation, one needs to view δx as a distribution
(introduced by L. Schwartz) and the equation should be interpreted in the
distributional derivative sense. The Dirac measure is the distributional limit
of the sequence of mollifiers, ρε , in the space of distributions.
6.8.2 Existence and Uniqueness
In this section, we shall give a formula for the solution of the Poisson equation
(6.8.1) in Rn in terms of the fundamental solution.
Theorem 6.8.4. For any given f ∈ Cc2 (Rn ), u := K ∗ f is a solution to the

Poisson equation (6.8.1).
Proof. By the property of convolution (cf. proof of Theorem F.0.10), we

know that Dα u(x) = (K ∗ Dα f )(x) for all |α| ≤ 2. Since f ∈ Cc2 (Rn ), we
have u ∈ C 2 (Rn ). The difficulty arises due to the singularity of K at the
origin. Consider, for any fixed m > 0,

Z
∆u(x) = K(y)∆x f (x − y) dy
Rn
Z Z
= K(y)∆x f (x − y) dy + K(y)∆x f (x − y) dy
Bm (0) Rn \Bm (0)
Z Z
= K(y)∆x f (x − y) dy + K(y)∆y f (x − y) dy
Bm (0) Rn \Bm (0)
Z Z
= K(y)∆x f (x − y) dy + K(y)∇y f (x − y) · ν dσy
Bm (0) Sm (0)
Z
− ∇y K(y) · ∇y f (x − y) dy (cf. Corollary D.0.7)
Rn \Bm (0)
Z Z
Bm (0) Sm (0)
Z
+ ∆y K(y)f (x − y) dy
Rn \Bm (0)
Z
− f (x − y)∇y K(y) · ν dσy (cf. Corollary D.0.7)
Sm (0)
Z Z
Bm (0) Sm (0)
Z
− f (x − y)∇y K(y) · ν dσy
Sm (0)
:= Im (x) + Jm (x) + Km (x).
But, due to the compact support of f , we have

Z
2
|Im (x)| ≤ kD f k∞,Rn |K(y)| dy.
Bm (0)
Thus, for n = 2,

m2 1
|Im (x)| ≤ + | ln m| kD2 f k∞,Rn
2 2
and for n ≥ 3, we have
m2
|Im (x)| ≤ kD2 f k∞,Rn .
2(n − 2)
Hence, as m → 0, |Im (x)| → 0. Similarly,

Z
|Jm (x)| ≤ |K(y)∇y f (x − y) · ν| dσy
Sm (0)
Z
≤ k∇f k∞,Rn |K(y)| dσy .
Sm (0)
Thus, for n = 2,
|Jm (x)| ≤ m| ln m|k∇f k∞,Rn
m
|Jm (x)| ≤ k∇f k∞,Rn .
(n − 2)
Hence, as m → 0, |Jm (x)| → 0. Now, to tackle the last term Km (x), we note
that a simple computation yields that ∇y K(y) = ωn−1 |y|n
y. Since we are in
the m radius sphere |y| = m. Also the unit vector ν outside of Sm (0), as a
boundary of Rn \ Bm (0), is given by −y/|y| = −y/m. Therefore,
1 1
∇y K(y) · ν = n+1
y·y = .
ωn m ωn mn−1
Z
Km (x) = − f (x − y)∇y K(y) · ν dσy
Sm (0)
Z
−1
= f (x − y) dσy
ωn mn−1 Sm (0)
Z
−1
= f (y) dσy
ωn mn−1 Sm (x)
Since f is continuous, for every ε > 0, there is a δ > 0 such that |f (x) −
f (y)| < ε whenever |x − y| < δ. When m → 0, we can choose m such that
m < δ and for this m, we see that Now, consider
Z
1
|Km (x) − (−f (x))| = f (x) − f (y) dσ y

ωn mn−1 Sm (x)
Z
1
= |f (x) − f (y)| dσy < ε.
ωn mn−1 Sm (x)
Thus, as m → 0, Km (x) → −f (x). Hence, u solves (6.8.1).
Remark 6.8.5. Notice that in the proof above, we have used the Green’s
identity eventhough our domain is not bounded (which is a hypothesis for
Green’s identity). This can be justified by taking a ball bigger than Bm (0)
and working in the annular region, and later letting the bigger ball approach
all of Rn .
A natural question at this juncture is: Is every solution of the Poisson

equation (6.8.1) of the form K ∗ f . We answer this question in the following
theorem.
Theorem 6.8.6. Let f ∈ Cc2 (Rn ) and n ≥ 3. If u is a solution of (6.8.1)

and u is bounded, then u has the form u(x) = (K ∗f )(x)+C, for any x ∈ Rn ,
where C is some constant.
Proof. We know that (cf. Theorem 6.8.4) u′ (x) := (K ∗ f )(x) solves (6.8.1),
the Poisson equation in Rn . Moreover, u′ is bounded for n ≥ 3, since K(x) →
0 as |x| → ∞ and f has compact support in Rn . Also, since u is given to be
a bounded solution of (6.8.1), v := u − u′ is a bounded harmonic function.
Hence, by Liouville’s theorem, v is constant. Therefore u = u′ + C, for some
constant C.
We turn our attention to studying Poisson equation in proper subsets of
R . Let Ω be an open bounded subset of Rn with C 1 boundary ∂Ω.
n
Theorem 6.8.7 (Uniqueness). Let Ω be an open bounded subset of Rn . For

the Poisson equation ∆u = f with one of Dirichlet, Robin or Mixed condi-
tions on ∂Ω, there exists at most one solution u ∈ C 2 (Ω) ∩ C 1 (Ω). In the
Neumann problem two solutions differ by a constant.
Proof. Let u and v be solutions of the Poisson equation with same boundary
conditions on ∂Ω. Then w := u − v is a harmonic function, ∆w = 0, with
homogeneous boundary condition on ∂Ω. By Green’s identity D.0.7, we have
Z Z
2
|∇w| dx = w(∇w · ν) dσ.
Ω ∂Ω
For the Drichlet, Neumann and Mixed case, the RHS is zero. For the Robin
condition the RHS is negative,
Z Z
w(∇w · ν) dσ = −c w2 dσ ≤ 0.
∂Ω ∂Ω
Thus, in all the four boundary conditions

Z
|∇w|2 dx ≤ 0
Ω
and ∇w = 0. Therefore, w = u − v is constant in the connected components

of Ω. In the case of Dirichlet, mixed and Robin the constant has to be zero,
by Maximum principle5 . Thus, u = v in these three cases.
To begin with we shall focus on the study of Dirichlet problem. The
Dirichlet problem is stated as follows: Given f : Ω → R and g : ∂Ω → R,
find u : Ω → R such that

−∆u = f in Ω
(6.8.2)
u = g on ∂Ω
Lemma 6.8.8. Let f be bounded and locally Hölder continuous6 with expo-
nent γ ≤ 1 in Ω. Then u := K ∗ f ∈ C 2 (Ω), −∆u = f in Ω.
Theorem 6.8.9 (Existence). Let Ω be a bounded domain with all boundary
points being regular w.r.t Laplacian. The classical Dirichlet problem (6.8.2) is
solvable (hence uniquely) for any bounded, locally Hölder continuous function
f in Ω and continuous function g on ∂Ω.
Proof. Recall that K is a fundamental solution of −∆. Set w(x) := f ∗ K in
Rn then −∆w = f . Set v = u − w. Then (6.8.2) is solvable iff

−∆v = 0 in Ω
v = g − w on ∂Ω
is solvable. The equation for v is solvable by Theorem 6.6.3.
6.8.3 Green’s Function

We shall now attempt to solve the Poisson equation on a proper open subset
Ω of Rn . This is done via the Green’s function. We begin by motivating the
Green’s function. For any x ∈ Ω, choose m > 0 such that Bm (x) ⊂ Ω. Set
Ωm := Ω \ Bm (x). Now applying the second identity of Corollary D.0.7 for
5
or, simply, from the fact that a non-zero c will contradict the continuous extension of
w to boundary.
6
Hölder continuous in each compact subset of Ω
any u ∈ C 2 (Ω) and vx (y) = K(y − x), the fundamental solution on Rn \ {x},
on the domain Ωm , we get
Z
u(y)∆y vx (y) dy
Ωm
Z Z
∂vx ∂u(y)
− vx (y)∆y u(y) dy = u(y) (y) − vx (y) dσy
Ωm ∂Ωm ∂ν ∂ν
Z Z
∂vx ∂u(y)
− vx (y)∆y u(y) dy = u(y) (y) − vx (y) dσy
Ωm ∂Ωm ∂ν ∂ν
Z Z Z Z
− + = +
Ω Bm (x) ∂Ω Sm (x)
Z
vx (y)∆y u(y) dy
Bm (x)
Z
∂vx
− u(y) (y) dσy
Sm (x) ∂ν
Z Z
∂u(y) ∂vx ∂u(y)
+ vx (y) dσy = u(y) (y) − vx (y) dσy
Sm (x) ∂ν ∂Ω ∂ν ∂ν
Z
+ vx (y)∆y u(y) dy
Z Ω
∂K ∂u(y)
Im (x) + Km (x) + Jm (x) = u(y) (y − x) − K(y − x) dσy
∂Ω ∂ν ∂ν
Z
+ K(y − x)∆y u(y) dy
Ω
The LHS is handled exactly as in the proof of Theorem 6.8.4, since u is

a continuous function on the compact set Ω and is bounded. We repeat the
arguments below for completeness sake. Consider the term Im .
Z
2
|Im (x)| ≤ kD uk∞,Ω |K(y − x)| dy.
Bm (x)
Thus, (
m2 1
2 2
+ | ln m| kD2 uk∞,Ω for n = 2
|Im (x)| ≤ m2
2(n−2)
kD2 uk∞,Ω for n ≥ 3.
Hence, as m → 0, |Im (x)| → 0. Next, consider the term Km (x). Note that
−1
∇y K(y−x) = ωn |y−x|n (y−x). Since we are in the m radius sphere |y−x| = m.
Also the unit vector ν inside of Sm (x), as a boundary of Ω \ Bm (x), is given

by −(y − x)/|y − x| = −(y − x)/m. Therefore,
1 1
∇y K(y − x) · ν = n+1
(y − x) · (y − x) = .
ωn m ωn mn−1
Thus,
Z
Km (x) = − u(y)∇y K(y − x) · ν dσy
Sm (x)
Z
−1
= u(y) dσy
ωn mn−1 Sm (x)
Since u is continuous, for every ε > 0, there is a δ > 0 such that |u(x)−u(y)| <
ε whenever |x − y| < δ. When m → 0, we can choose m such that m < δ
and for this m, we see that Now, consider
Z
1
|Km (x) − (−u(x))| = u(x) − u(y) dσ y

ωn mn−1 Sm (x)
Z
1
= |u(x) − u(y)| dσy < ε.
ωn mn−1 Sm (x)
Thus, as m → 0, Km (x) → −u(x). Finally, we consider the term Jm (x),

Z
|Jm (x)| ≤ |K(y − x)∇y u(y) · ν| dσy
Sm (x)
Z
≤ k∇y uk∞,Ω |K(y − x)| dσy .
Sm (x)
Thus, for n = 2,
(
m| ln m|k∇y uk∞,Ω for n = 2
|Jm (x)| ≤ m
|Jm (x)| ≤ (n−2) k∇y uk∞,Ω for n ≥ 3.
Hence, as m → 0, |Jm (x)| → 0. Therefore, letting m → 0, we have the

identity
Z Z
∂u(y) ∂K
u(x) = K(y − x) − u(y) (y − x) dσy − K(y−x)∆y u(y) dy
∂Ω ∂ν ∂ν Ω
(6.8.3)
For the Dirichlet problem, ∆u is known in Ω and u is known on ∂Ω.

Thus, (6.8.3) gives an expression for the solution u, provided we know the
normal derivative ∂u(y)
∂ν
along ∂Ω. But this quantity is usually an unknown for
Dirichlet problem. Thus, we wish to rewrite (6.8.3) such that the knowledge
of the normal derivative is not necessary. To do so, we introduce a function
ψx (y), for a fixed x ∈ Ω, as the solution of the boundary-value problem,

∆ψx (y) = 0 in Ω
(6.8.4)
ψx (y) = K(y − x) on ∂Ω.
Now applying the second identity of Corollary D.0.7 for any u ∈ C 2 (Ω)
and v(y) = ψx (y), we get
Z Z
∂ψx ∂u
u − ψx dσy = (u∆y ψx − ψx ∆y u) dy.
∂Ω ∂ν ∂ν Ω
Therefore, substituting the following identity

Z Z Z
∂u(y) ∂ψx (y)
K(y − x) dσy = ψx (y)∆y u(y) dy + u(y) dσy
∂Ω ∂ν Ω ∂Ω ∂ν
in (6.8.3), we get
Z Z
u(x) = (ψx (y) − K(y − x)) ∆y u dy + u∇ (ψx (y) − K(y − x)) · ν dσy .
Ω ∂Ω
The identity above motivates the definition of what is called the Green’s
function.
Definition 6.8.10. For any given open subset Ω ⊂ Rn and x, y ∈ Ω such

that x 6= y, we define the Green’s function as
G(x, y) := ψx (y) − K(y − x).
Rewriting (6.8.3) in terms of Green’s function,we get

Z Z
∂G(x, y)
u(x) = G(x, y)∆y u(y) dy + u(y) dσy .
Ω ∂Ω ∂ν
Thus, in the arguments above we have proved the following theorem.

Theorem 6.8.11. Let Ω be a bounded open subset of Rn with C 1 boundary.

Also, given f ∈ C(Ω) and g ∈ C(Ω). If u ∈ C 2 (Ω) solves the Dirichlet
problem (6.8.2), then u has the representation
Z Z
∂G(x, y)
u(x) = − G(x, y)f (y) dy + g(y) dσy . (6.8.5)
Ω ∂Ω ∂ν
Observe that we have solved the Dirichlet problem (6.8.2) provided we

know the Green’s function. The construction of Green’s function depends
on the construction of ψx for every x ∈ Ω. In other words, (6.8.2) is solved
if we can solve (6.8.4). Ironically, computing ψx is usually possible when Ω
has simple geometry. We shall identify two simple cases of Ω, half-space and
balll, where we can explicitly compute G.
The Green’s function is the analogue of the fundamental solution K for
the boundary value problem. This is clear by observing that, for a fixed
x ∈ Ω, G satisfies (informally) the equation,

−∆G(x, ·) = δx in Ω
G(x, ·) = 0 on ∂Ω,
where δx is the Dirac measure at x.
Theorem 6.8.12. For all x, y ∈ Ω such that x 6= y, we have G(x, y) =

G(y, x), i.e., G is symmetric in x and y.
Proof. Let us fix x, y ∈ Ω. For a fixed m > 0, set Ωm = Ω \ (Bm (x) ∪ Bm (y))
and applying Green’s identity for v(·) := G(x, ·) and w(·) := G(y, ·), we get
Z Z
∂w(z) ∂v(z)
v(z) − w(z) dσz = v(z)∆z w(z) dz
∂Ωm ∂ν ∂ν Ωm
Z
− w(z)∆z v(z) dz
Ωm
Z
∂w(z) ∂v(z)
v(z) − w(z) dσz = 0
∂Ωm ∂ν ∂ν
Z Z
∂w ∂v ∂v ∂w
v −w dσz = w −v dσz
Sm (x) ∂ν ∂ν Sm (y) ∂ν ∂ν
Jm (x) + Km (x) = Jm (y) + Km (y).
Z
|Jm (x)| ≤ |v(z)∇z w(z) · ν| dσz
Sm (x)
Z
≤ k∇wk∞,Ω |v(z)| dσz
Sm (x)
Z
= k∇wk∞,Ω |ψx (z) − K(z − x)| dσz .
Sm (x)
Thus, for n = 2,
|Jm (x)| ≤ (2πmkψx k∞,Ω + m| ln m|) k∇wk∞,Ω

n−1 m
|Jm (x)| ≤ ωn m kψx k∞,Ω + k∇wk∞,Ω .
(n − 2)
Hence, as m → 0, |Jm (x)| → 0. Now, consider the term Km (x),
Z
∂v(z)
Km (x) = − w(z) dσz
Sm (x) ∂ν
Z Z
∂K ∂ψx (z)
= w(z) (z − x) dσz − w(z) dσz .
Sm (x) ∂ν Sm (x) ∂ν
The second term goes to zero by taking the sup-norm outside the integral.
−1
To tackle the first term, we note that ∇z K(z − x) = ωn |z−x| n (z − x). Since
we are in the m radius sphere |z − x| = m. Also the unit vector ν outside of

Sm (x), as a boundary of Ω\Bm (x), is given by −(z −x)/|z −x| = −(z −x)/m.
Therefore,
1 1
∇z K(z − x) · ν = (z − x) · (z − x) = .
ωn mn+1 ωn mn−1
Z Z
1
w(z)∇z K(z − x) · ν dσz = w(z) dσz
Sm (x) ωn mn−1 Sm (x)
Since w is continuous in Ω \ {y}, for every ε > 0, there is a δ > 0 such that
|w(z) − w(x)| < ε whenever |x − z| < δ. When m → 0, we can choose m
such that m < δ and for this m, we see that Now, consider
Z
1
w(z) dσ − w(x)
ωn mn−1 z
Sm (x)
Z
1
= |w(z) − w(x)| dσz < ε.
ωn mn−1 Sm (x)
Thus, as m → 0, Km (x) → w(x). Arguing similarly, for Jm (y) and Km (y),

we get G(y, x) = G(x, y).
6.8.4 Green’s Function for half-space

In this section, we shall compute explicitly the Green’s function for positive
half-space. Thus, we shall have
Rn+ = {x = (x1 , x2 , . . . , xn ) ∈ Rn | xn > 0}
and
∂Rn+ = {x = (x1 , x2 , . . . , xn } ∈ Rn | xn = 0}.
To compute the Green’s function, we shall use the method of reflection.
The reflection technique ensures that the points on the boundary (along
which the reflection is done) remains unchanged to respect the imposed
Dirichlet condition.
Definition 6.8.13. For any x = (x1 , x2 , . . . , xn ) ∈ Rn+ , we define its reflec-

tion along its boundary Rn−1 as x⋆ = (x1 , x2 , . . . , −xn ).
It is obvious from the above definition that, for any y ∈ ∂Rn+ , |y − x⋆ | =

|y − x|. Given a fixed x ∈ Rn+ , we need to find a harmonic function ψx in
Rn+ , as in (6.8.4). But K(· − x) is harmonic in Rn+ \ {x}. Thus, we use the
method of reflection to shift the singularity of K from Rn+ to the negative
half-space and define
ψx (y) = K(y − x⋆ ).
By definition, ψx is harmonic in Rn+ and on the boundary ψx (y) = K(y − x).
Therefore, we define the Green’s function to be G(x, y) = K(y − x⋆ ) − K(y −
x), for all x, y ∈ Rn+ and x 6= y. It now only remains to compute the normal
−1
derivative of G. Recall that ∇y K(y − x) = ωn |y−x| n (y − x). Thus,

−1 y − x⋆ y−x
∇y G(x, y) = ⋆ n
−
ωn |y − x | |y − x|n
Therefore, when y ∈ ∂Rn+ , we have
−1
∇y G(x, y) = (x − x⋆ ).
ωn |y − x|n
Since the outward unit normal of ∂Rn+ is ν = (0, 0, . . . , 0, −1), we get
2xn
∇y G(x, y) · ν = .
ωn |y − x|n
Definition 6.8.14. For all x ∈ Rn+ and y ∈ ∂Rn+ , the map
2xn
P (x, y) :=
ωn |y − x|n
is called the Poisson kernel for Rn+ .
Now substituing for G in (6.8.5), we get the Poisson formula for u,

Z Z
⋆ 2xn g(y)
u(x) = [K(y − x) − K(y − x )]f (y) dy + dσy . (6.8.6)
Rn
+
ωn ∂Rn
+
|y − x|n
It now remains to show that the u as defined above is, indeed, a solution of
(6.8.2) for Rn+ .
Exercise 30. Let f ∈ C(Rn+ ) be given. Let g ∈ C(Rn−1 ) be bounded. Then
u as given in (6.8.6) is in C 2 (Rn+ ) and solves (6.8.2).
6.8.5 Green’s Function for a disk

In this section, we shall compute explicitly the Green’s function for a ball of
radius r > 0 and centred at a ∈ Rn , Br (a). As usual, we denote the surface
of the disk as Sr (a), the circle of radius r centred at a. We, once again, use
the method of reflection but, this time reflected along the boundary of the
disk.
Definition 6.8.15. For any x ∈ Rn \ {a}, we define its reflection along the
2 (x−a)
circle Sr (a) as x⋆ = r|x−a|2 + a.
The idea behind reflection is clear for the unit disk, i.e., when a = 0 and
r = 1, as x⋆ = |x|x2 . The above definition is just the shift of origin to a and
dilating the unit disk by r.
Now, for any y ∈ Sr (a) and x 6= a, consider
|y − x⋆ |2 = |y − a|2 − 2(y − a) · (x⋆ − a) + |x⋆ − a|2

2
x−a r (x − a) 2
2 2
= r − 2r (y − a) · +
|x − a|2 |x − a|2
r2
= 2
(|x − a|2 − 2(y − a) · (x − a) + r2 )
|x − a|
r2
= (|x − a|2 − 2(y − a) · (x − a) + |y − a|2 )
|x − a|2
r2
= |y − x|2
|x − a|2
Therefore, |x−a|
r
|y−x⋆ | = |y−x| for all y ∈ Sr (a). For each fixed x ∈ Br (a), we
need to find a harmonic function ψx in Br (a) solving (6.8.4). Since K(· − x)
is harmonic in Br (a) \ {x}, we use the method of reflection to shift the
singularity of K at x to the complement of Br (a). Thus, we define

|x − a| ⋆
ψx (y) = K (y − x ) x 6= a.
r
2−n
For n ≥ 3, K |x−a| r
(y − x ⋆
) = |x−a|
r 2−n
K(y − x⋆ ). Thus, for n ≥ 3, ψx solves
(6.8.4), for x 6= a. For n = 2,

|x − a| ⋆ −1 |x − a|
K (y − x ) = ln + K(y − x⋆ ).
r 2π r
Hence ψx solves (6.8.4) for n = 2. Note that we are yet to identify a harmonic
function ψa corresponding to x = a. We do this by setting ψa to be the
constant function (
1
− 2π ln r (n = 2)
ψa (y) := r 2−n
ωn (n−2)
(n ≥ 3).
Thus, ψa is harmonic and solves (6.8.4) for x = a. Therefore, we define the
Green’s function to be

|x − a| ⋆
G(x, y) := K (y − x ) − K(y − x) ∀x, y ∈ Br (a), x 6= a and x 6= y
r
and (
1 r
− 2π ln |y−a|
(n = 2)
G(a, y) :=
1 2−n
ωn (n−2)
(r − |y − a|2−n ) (n ≥ 3).
We shall now compute the normal derivative of G. Recall that
−1
∇y K(y − x) = (y − x)
ωn |y − x|n

−|x−a|2−n
and one can compute ∇y K |x−a|
r
(y − x ⋆
) = r2−n ωn |y−x⋆ |n
(y−x⋆ ). Therefore,

−1 |x − a|2−n (y − x⋆ ) y−x
∇y G(x, y) = − .
ωn r2−n |y − x⋆ |n |y − x|n
If y ∈ Sr (a), we have

−1 |x − a|2 ⋆
∇y G(x, y) = (y − x ) − (y − x)
ωn |y − x|n r2

−1 |x − a|2
= − 1 (y − a)
ωn |y − x|n r2
Since the outward unit normal at any point y ∈ Sr (a) is 1r (y − a), we have
X n
−1 |x − a|2 1
∇y G(x, y) · ν = n 2
−1 (yi − ai )2
ωn |y − x| r r
2
i=1
−r |x − a|
= n
−1 .
ωn |y − x| r2
Definition 6.8.16. For all x ∈ Br (a) and y ∈ Sr (a), the map
r2 − |x − a|2
P (x, y) :=
rωn |y − x|n
is called the Poisson kernel for Br (a).
Now substituing for G in (6.8.5), we get the Poisson formula for u,
Z Z
r2 − |x − a|2 g(y)
u(x) = − G(x, y)f (y) dy + n
dσy . (6.8.7)
Br (a) rωn Sr (a) |y − x|
It now remains to show that the u as defined above is, indeed, a solution of
(6.8.2) for Br (a).
Exercise 31. Let f ∈ C(Br (a)) be given. Let g ∈ C(Sr (a)) be bounded.
Then u as given in (6.8.7) is in C 2 (Br (a)) and solves (6.8.2).
6.8.6 Conformal Mapping and Green’s Function

In two dimensions, the Green’s function has a nice connection with conformal
mapping. Let w = f (z) be a conformal mapping from an open domain
(connected) Ω ⊂ R2 onto the interior of the unit circle. The Green’s function
of Ω is

1 1 − f (z)f (z )
0
G(z, z0 ) = ln
2π f (z) − f (z0 )
where z = x1 + ix2 and z0 = y1 + iy2 .
6.8.7 Dirichlet Principle

The Dirichlet principle (formulated, independently by Gauss, Lord Kelvin
and Dirichlet) states that the solution of the Dirichlet problem minimizes
the corresponding energy functional.
Let Ω be an open bounded subset of Rn with C 1 boundary ∂Ω and let
f : Ω → R and g : ∂Ω → R be given. For convenience, recall the Dirichlet
problem ((6.8.2)),

−∆u = f in Ω
u = g on ∂Ω.
Any solution u of (6.8.2) is in V = {v ∈ C 2 (Ω) | v = g on ∂Ω}. The

energy functional J : V → R is defined as
Z Z
1 2
J(v) := |∇v| dx − f v dx
2 Ω Ω
Theorem 6.8.17 (Dirichlet’s principle). A C 2 (Ω) function u solves (6.8.2)

iff u minimises the functional J on V , i.e.,
J(u) ≤ J(v) ∀v ∈ V.
Proof. Let u ∈ C 2 (Ω) be a solution of (6.8.2). For any v ∈ V , we multiply

both sides of (6.8.2) by u − v and integrating we get,

Z Z
(−∆u)(u − v) dx = f (u − v) dx
Ω Ω
Z Z
∇u · ∇(u − v) dx = f (u − v) dx
Ω Ω
Z Z
2

|∇u| − f u dx = (∇u · ∇v − f v) dx
Ω
ZΩ Z
≤ |∇u · ∇v| − f v dx
Ω Ω
Z Z
1 2 2

≤ |∇u| + |∇v| dx − f v dx
2 Ω Ω
(since 2ab ≤ a2 + b2 )
J(u) ≤ J(v).
Thus, u minimises J in V . Conversely, let u minimise J in V . Thus,
J(u) ≤ J(v) ∀v ∈ V
J(u) ≤ J(u + tφ) (for any φ ∈ C 2 (Ω) such that φ = 0 on ∂Ω)
1
0 ≤ (J(u + tφ) − J(u))
t Z Z
1 1 2 2

0 ≤ t |∇φ| + 2t∇φ · ∇u dx − t f φ dx
t 2 Ω Ω
Taking limit t → 0 both sides, we get

Z Z
0≤ ∇φ · ∇u dx − f φ dx ∀φ ∈ C 2 (Ω) s.t. φ = 0 on ∂Ω.
Ω Ω
Choosing −φ in place of φ we get the reverse inequality, and we have equality

in the above. Thus,
Z Z
∇u · ∇φ dx = f φ dx ∀φ ∈ C 2 (Ω) s.t. φ = 0 on ∂Ω
Ω Ω
Z
(−∆u − f )φ dx = 0 ∀φ ∈ C 2 (Ω) s.t. φ = 0 on ∂Ω.
Ω
Thus u solves (6.8.2).

6.9 Neumann Boundary Condition

The Neumann problem is stated as follows: Given f : Ω → R and g : ∂Ω →
R, find u : Ω → R such that

−∆u = f in Ω
∂u (6.9.1)
∂ν
= g on ∂Ω
where ∂u
∂ν
:= ∇u · ν and ν = (ν1 , . . . , νn ) is the outward pointing unit normal
vector field of ∂Ω. Thus, the boundary imposed is called the Neumann
boundary condition. The solution of a Neumann problem is not necessarily
unique. If u is any solution of (6.9.1), then u + c for any constant c is also a
solution of (6.9.1). More generally, for any v such that v is constant on the
connected components of Ω, u + v is a solution of (6.9.1).
Appendices
171
Appendix A
The Gamma Function
The gamma function Γ : (0, ∞) → R is defined as,

Z ∞
Γ(x) = e−t tx−1 dt ∀x ∈ (0, ∞).
0
Note that the gamma function is defined as an improper integral and its
existence has to be justified. Observe that for a fixed x > 0,
|e−t tx−1 | = e−t tx−1 ≤ tx−1 ∀t > 0,
R1
since for t > 0, |e−t | ≤ 1. Now, since 0 tx−1 dt exists, we have by comparison
R1
test the existence of the integral 0 e−t tx−1 dt. Now, for t → ∞, e−t tx−1 → 0
and hence the is a constant C > 0 such that
tx−1 e−t ≤ C/t2 ∀t ≥ 1.
R∞
Since 1 1/tR2 dt exists, we again have using comparison test the existence of
∞
the integral 1 e−t tx−1 dt. In fact, the gamma function can be defined for
any complex number z ∈ C such that Re(z) > 0.
It is worth noting that the gamma function Γ generalises the notion of
factorial of positive integers. This would be the first property we shall prove.
Exercise 32. Show that Γ(x + 1) = xΓ(x). In particular, for any positive
√
integer n, Γ(n + 1) = n!. Also, show that Γ(1) = 1 and Γ(1/2) = π.
Further, for any positive integer n,
√
Γ(n + 1/2) = (n − 1/2)(n − 3/2) . . . (1/2) π.
(Hint: Use integration by parts and change of variable).
173
APPENDIX A. THE GAMMA FUNCTION 174
Exercise 33. Show that Γ is continuous on (0, ∞).

Exercise 34. Show that the logarithm of Γ is convex on (0, ∞).
We shall now show that Γ is the only possible generalisation of the notion
of factorial of positive integers satisfying the above properties.
Theorem A.0.1. Let f be positive and continuous on (0, ∞) and let log f
be convex on (0, ∞). Also, let f satisfy the recursive equation
f (x + 1) = xf (x) ∀x > 0
and f (1) = 1, then f (x) = Γ(x) for all x > 0.

Appendix B
Normal Vector of a Surface
Let S(x, y, z) = 0 be the equation of a surface S in R3 . Let us a fix a

point p0 = (x0 , y0 , z0 ) ∈ S. We need to find the normal vector at p0 for
the surface S. Let us fix an arbitrary curve C lying on the surface passing
through the point p0 . Let the parametrized form of the curve C be given as
r(t) = (x(t), y(t), z(t)) such that r(t0 ) = p0 . Since the curve C ≡ r(t) lies
on the surface for all t, we have S(r(t)) = 0. Thus, S(x(t), y(t), z(t)) = 0.
Differentiating w.r.t t (using chain rule), we get
∂S dx(t) ∂S dy(t) ∂S dz(t)

+ + = 0
∂x dt ∂y dt ∂z dt
(Sx , Sy , Sz ) · (x′ (t), y ′ (t), z ′ (t)) = 0
∇S(r(t)) · r′ (t) = 0.
In particular, the above computation is true for the point p0 . Since r′ (t0 )
is the slope of the tangent at t0 to the curve C, we see that the vector ∇S(p0 )
is perpendicular to the tangent vector at p0 . Since this argument is true for
any curve that passes through p0 . We have that ∇S(p0 ) is normal vector to
the tangent plane at p0 . If, in particular, the equation of the surface is given
as S(x, y, z) = u(x, y) − z, for some u : R2 → R, then
∇S(p0 ) = (Sx (p0 ), Sy (p0 ), Sz (p0 ))

= (ux (x0 , y0 ), uy (x0 , y0 ), −1) = (∇u(x0 , y0 ), −1).
175
APPENDIX B. NORMAL VECTOR OF A SURFACE 176
Appendix C
Implicit Function Theorem
Theorem C.0.2 (Implicit Function Theorem). Let Ω ⊂ Rm × Rn be an open

subset and f : Ω → Rn such that f is continuously differentiable (C 1 ) in Ω.
Let (x0 , y0 ) ∈ Ω be such that f (x0 , y0 ) = 0 and the n × n matrix
 ∂f1 ∂f1 
∂y1
(x0 , y0 ) · · · ∂y n
(x0 , y0 )
 .. ... .. 
Dy f (x0 , y0 ) :=  . . 
∂fn ∂fn
∂y1
(x0 , y0 ) ··· ∂yn
(x0 , y0 )
is non-singular, then there is a neighbourhood U ⊂ Rm of x0 and a unique

function g : U → Rn such that g(x0 ) = y0 and, for all x ∈ U , f (x, g(x)) = 0.
Further g is continuously differentiable in U .
177
APPENDIX C. IMPLICIT FUNCTION THEOREM 178
Appendix D
Divergence Theorem
Definition D.0.3. For an open set Ω ⊂ Rn we say that its boundary ∂Ω is

C k (k ≥ 1), if for every point x ∈ ∂Ω, there is a r > 0 and a C k diffeomor-
phism γ : Br (x) → B1 (0) ( i.e. γ −1 exists and both γ and γ −1 are k-times
continuously differentiable) such that
1. γ(∂Ω ∩ Br (x)) ⊂ B1 (0) ∩ {x ∈ Rn | xn = 0} and
2. γ(Ω ∩ Br (x)) ⊂ B1 (0) ∩ {x ∈ Rn | xn > 0}
We say ∂Ω is C ∞ if ∂Ω is C k for all k = 1, 2, . . . and ∂Ω is analytic if γ is

analytic.
Equivalently, a workable definition of C k boundary would be the follow-

ing: if for every point x ∈ ∂Ω, there exists a neighbourhood Bx of x and a
C k function γ : Rn−1 → R such that
Ω ∩ Bx = {x ∈ Bx | xn > γ(x1 , x2 , . . . , xn−1 )}.
The divergence of a vector field is the measure of the magnitude (outgoing

nature) of all source (of the vector field) and absorption in the region. The
divergence theorem was discovered by C. F. Gauss in 18131 which relates
the outward flow (flux) of a vector field through a closed surface to the
behaviour of the vector field inside the surface (sum of all its “source” and
“sink”). The divergence theorem is, in fact, the mathematical formulation
of the conservation law.
1
J. L. Lagrange seems to have discovered this, before Gauss, in 1762
179
APPENDIX D. DIVERGENCE THEOREM 180
Theorem D.0.4. Let Ω be an open bounded subset of Rn with C 1 boundary.

If v ∈ C 1 (Ω) then Z Z
∂v
dx = vνi dσ
Ω ∂xi ∂Ω
where ν = (ν1 , . . . , νn ) is the outward pointing unit normal vector field and
dσ is the surface measure of ∂Ω.

∂v
The domain Ω need not be bounded provided |v| and ∂x i
decays as
|x| → ∞. The field of geometric measure theory attempts to identify the
precise condition on ∂Ω and v for which divergence theorem or integration
by parts hold.
Corollary D.0.5 (Integration by parts). Let Ω be an open bounded subset
of Rn with C 1 boundary. If u, v ∈ C 1 (Ω) then
Z Z Z
∂v ∂u
u dx + v dx = uvνi dσ.
Ω ∂xi Ω ∂xi ∂Ω
Theorem D.0.6 (Gauss). Let Ω be an open bounded subset of Rn with C 1

boundary. Given a vector field V = (v1 , . . . , vn ) on Ω such that vi ∈ C 1 (Ω)
for all 1 ≤ i ≤ n, then
Z Z
∇ · V dx = V · ν dσ. (D.0.1)
Ω ∂Ω
Corollary D.0.7 (Green’s Identities). Let Ω be an open bounded subset of

Rn with C 1 boundary. Let u, v ∈ C 2 (Ω), then
(i) Z Z
∂u
(v∆u + ∇v · ∇u) dx = v dσ,
Ω ∂Ω ∂ν
∂u
where ∂ν
:= ∇u · ν and ∆ := ∇ · ∇.
(ii) Z Z
∂u ∂v
(v∆u − u∆v) dx = v −u dσ.
Ω ∂Ω ∂ν ∂ν
Proof. Apply divergence theorem to V = v∇u to get the first formula. To get
second formula apply divergence theorem for both V = v∇u and V = u∇v
and subtract one from the other.
Appendix E
Surface Area and Volume of

Disk in Rn
Theorem E.0.8 (Polar coordinates). Let f : Rn → R be continuous and

integrable. Then
Z Z ∞ Z
f (x) dx = f (y) dσy dr
Rn 0 Sr (a)
for each a ∈ Rn . In particular, for each r > 0,

Z Z
d
f (x) dx = f (y) dσy .
dr Br (a) Sr (a)
Theorem E.0.9. Prove that

Z
2
e−π|x| dx = 1.
Rn
Further, prove that the surface area ωn of S1 (0) in Rn is
2π n/2
Γ(n/2)
and the volume of the ball B1 (0) in Rn is ωn /n. Consequently, for any x ∈ Rn
and the r > 0, the surface area of Sr (x) is rn−1 ωn and the volume of Br (x)
is rn ωn /n.
181
APPENDIX E. SURFACE AREA AND VOLUME OF DISK IN RN 182
Proof. We first observe that

Pn n
Y
e−π|x| = e−π( x2i )
2 2
i=1 = e−πxi .
i=1
Therefore,
Z Z n
Y
−π|x|2 2
In := e dx = e−πxi dx
Rn R i=1n
YZ
n
2
= e−πt dt
i=1 R
Z n
−πt2
= e dt = (I1 )n
R
Z Z 2(n/2)
−π|x|2 −πt2
n/2
e dx = e dt = (I1 )2 = (I2 )n/2
Rn R
Z n/2
−π|y|2
= e dy
R2
Z 2π Z ∞ n/2
−π|y|2
= e dy
0 0
Z 2π Z ∞ n/2
−πr 2
= e r dr dθ (since jacobian is r)
0 0
Z ∞ n/2
−πr 2
= 2π e r dr
0
Z ∞ n/2
−πs
= π e ds (by setting r2 = s)
0
Z ∞ n/2
−q
= e dq (by setting πs = q)
0
= (Γ(1))n/2 = 1.
Let ωn denote the surface area of the unit sphere S1 (0) in Rn , i.e.,
Z
ωn = dσ,
S1 (0)
where dσ is the n − 1-dimensional surface measure. Now, consider

Z
2
1 = e−π|x| dx
n
ZR Z ∞
2
= e−πr rn−1 dr dσ
S1 (0) 0
Z ∞
2
= ωn e−πr rn−1 dr
0
Z ∞
ωn
= e−s s(n/2)−1 ds (by setting s = πr2 )
2π n/2 0
ωn Γ(n/2)
= .
2π n/2
2π n/2
Thus, ωn = Γ(n/2)
. We shall now compute the volume of the disk B1 (0).
Consider, Z Z 1
ωn
dx = ωn rn−1 dr = .
B1 (0) 0 n
n
For any x ∈ R and r > 0, we observe by the shifting of origin that the
surface area of Sr (x) is same as the surface area of Sr (0). Let Sr (0) = {s ∈
Rn | |s| = r}. Now
Z Z
dσs = rn−1 dσt = rn−1 ωn ,
Sr (0) S1 (0)
where t = s/r. Thus, the surface area of Sr (x) is rn−1 ωn . Similarly, volume
of a disk Br (x) is rn ωn /n.
Appendix F
Mollifiers and Convolution
Exercise 35. Show that the Cauchy’s exponential function, f : R → R defined

as (
exp(−x−2 ) if x > 0
f (x) =
0 if x ≤ 0,
is infinitely differentiable, i.e., is in C ∞ (R).
Using the above Cauchy’s exponential function, one can construct func-
tions in Cc∞ (Rn ).
Exercise 36. For x = (x1 , . . . , xn ) ∈ Rn , show that ρ : Rn → R defined as
(
−1
cexp( 1−|x| 2) if |x| < 1
ρ(x) =
0 if |x| ≥ 1,
is in CRc∞ (Rn ) with supp(ρ) = B1 (0), ball with centre 0 and radius 1, where
−1
c−1 = |x|≤1 exp( 1−|x| 2 ) dx.
Thus, one can introduce a sequence of functions in Cc∞ (Rn ), called mol-
lifiers. For ε > 0, we set
( 2
cε−n exp( ε2−ε
−|x|2
) if |x| < ε
ρε (x) = (F.0.1)
0 if |x| ≥ ε,
Equivalently, ρε (x) = ε−n ρ(x/ε).

R
Exercise 37. Show that ρε ≥ 0 and Rn
ρε (x) dx = 1 and is in Cc∞ (Rn ) with
support in Bε (0).
185
APPENDIX F. MOLLIFIERS AND CONVOLUTION 186
Let f, g ∈ L1 (Rn ). Their convolution f ∗ g is defined as, for x ∈ Rn ,

Z
(f ∗ g)(x) = f (x − y)g(y) dy.
Rn
The integral on RHS is well-defined, since by Fubini’s Theorem and the

translation invariance of the Lebesgue measure, we have
Z Z Z Z
|f (x−y)g(y)| dx dy = |g(y)| dy |f (x−y)| dx = kgk1 kf k1 < ∞.
Rn Rn Rn Rn
Thus, for a fixed x, f (x − y)g(y) ∈ L1 (Rn ).
Theorem F.0.10. Let Ω ⊂ Rn be an open subset of Rn and let
Ωε := {x ∈ Ω | dist(x, ∂Ω) > ε}.
RIf u : Ω → R is locally integrable, i.e.,

∞
for every compact subset K ⊂ Ω,
K
|u| < +∞, then uε := ρε ∗ u is in C (Ωε ).
Proof. Fix x ∈ Ωε . Consider

Z
uε (x + hei ) − uε (x) 1
= [ρε (x + hei − y) − ρε (x − y)] u(y) dy
h h Ω
Z
1
= [ρε (x + hei − y) − ρε (x − y)]u(y) dy.
Bε (x) h
Now, taking limh→0 both sides, we get

Z
∂uε (x) 1
= lim [ρε (x + hei − y) − ρε (x − y)]u(y) dy
∂xi h→0 B (x) h
ε
Z
∂ρε (x − y)
= u(y) dy
Bε (x) ∂xi
(interchange of limits is due to the uniform convergence)
Z
∂ρε (x − y) ∂ρε
= u(y) dy = ∗ u.
Ω ∂xi ∂xi
Similarly, one can show that, for any tuple α, Dα uε (x) = (Dα ρε ∗ u)(x).
Thus, uε ∈ C ∞ (Ωε ).
Appendix G
Duhamel’s Principle
Consider the first order inhomogeneous ODE

′
x (t) + ax(t) = f (t) in (0, ∞)
(G.0.1)
x(0) = x0 .
Multiplying the integration factor eat both sides, we get
[eat x(t)]′ = eat f (t)
and Z t
−at
x(t) = e eas f (s) ds + ce−at .
0
Using the initial condition x(0) = x0 , we get
Z t
−at
x(t) = x0 e + ea(s−t) f (s) ds.
0
Notice that x0 e−at is a solution of the homogeneous ODE. Thus, the solution
x(t) can be given as
Z t
x(t) = S(t)x0 + S(t − s)f (s) ds
0
where S(t) is a solution operator of the linear equation, given as S(t) = e−at .
Consider the second order inhomogeneous ODE
 ′′
 x (t) + a2 x(t) = f (t) in (0, ∞)
x(0) = x0 (G.0.2)
 ′
x (0) = x1 .
187
APPENDIX G. DUHAMEL’S PRINCIPLE 188
We introduce a new function y such that
x′ (t) = ay(t).
Then
f (t)
y ′ (t) = − ax(t)
a
and the second order ODE can be rewritten as a system of first order ODE
X ′ (t) + AX(t) = F (t)
where X = (x, y), F = (0, f /a) and

0 −a
A=
a 0
with the initial condition X0 := X(0) = (x0 , x1 /a). We introduce the matrix
P (At)n
exponential eAt = ∞ n=1 n! . Then, multiplying the integration factor e
At
both sides, we get

[eAt X(t)]′ = eAt F (t)
and Z t
−At
X(t) = X0 e + eA(s−t) F (s) ds.
0
Notice that X0 e−At is a solution of the homogeneous ODE. Thus, the solution
X(t) can be given as
Z t
X(t) = S(t)X0 + S(t − s)F (s) ds
0
where S(t) is a solution operator of the linear equation, given as S(t) = e−At .
Bibliography
189
BIBLIOGRAPHY 190
Index
ball Laplace operator, 3, 12, 117

surface area of, 181 Liouville’s theorem, 137
volume of, 181
maximum principle
characteristic equations, 25 strong, 133
convolution, 131, 186 weak, 134
mean value property, 130
Dirac measure, 152, 161 method of characteristics, 24
directional derivative, 2 method of reflection, 163, 164
Dirichlet Problem, 157 mollifiers, 185
elliptic PDE, 63, 66 Monge cone, 23
energy functional, 167 Neumann boundary condition, 169
function Neumann Problem, 169
Cauchy’s exponential, 185
parabolic PDE, 63, 65
fundamental solution, 150, 152
Poisson equation, 12, 150
gamma function, 173 Poisson formula, 164, 166
Gauss divergence result, 180 Poisson kernel, 164, 166
gradient, 2
radial function, 118
Green’s function, 160
Green’s identities, 180 transport equation, 20
tricomi equation, 68
Hadamard, 9
Hadamard’s wellposedness, 9
harmonic function, 12, 124
Harnack inequality, 137
Hessian matrix, 3, 62
hyperbolic PDE, 63, 65
integral curve, 22
integral surface, 23
191

Partial Differential Equations: T. Muthukumar Tmk@iitk - Ac.in April 18, 2014

Uploaded by

Partial Differential Equations: T. Muthukumar Tmk@iitk - Ac.in April 18, 2014

Uploaded by

Partial Differential Equations

April 18, 2014

2 First Order PDE 17

3.1.1 Semilinear: Two Dimension . . . . . . . . . . . . . . . 63

5 Heat Equation 107

6 The Laplacian 117

6.4.2 Properties of Harmonic Functions . . . . . . . . . . . . 130

A The Gamma Function 173

B Normal Vector of a Surface 175

C Implicit Function Theorem 177

D Divergence Theorem 179

E Surface Area and Volume of Disk in Rn 181

F Mollifiers and Convolution 185

G Duhamel’s Principle 187

Ω denotes an open subset of Rn , not necessarily bounded

∂Ω denotes the boundary of Ω

C k (X) is the class of all k-times (k ≥ 1) continuously differentiable functions

C ∞ (X) is the class of all infinitely differentiable functions on X

Br (x) denotes the open disk with centre at x and radius r

Sr (x) denotes the circle or sphere with centre at x and radius r

wn denotes the surface area of a n-dimensional sphere of radius 1.

A partial differential equation (PDE) is an equation involving an unknown

(ii) Solving the equation (PDE). What do we mean by a solution of the

(iii) Studying properties of the solution. Usually, we do not end up with a

The partial differential operator of order α is denoted as,

k = 2, we regard D2 as being arranged in a matrix form (called the Hessian

∇u = (ux , uy ) = (2ax, 2by).

for each x ∈ Ω and u : Ω → R is the unknown.

• if u is k-times differentiable with the k-th derivative being continuous

• and u satisfies the equation (1.1.1).

uxxxx + uyyyy + 2uxxyy = 0.

Exercise 2. Determine the relation between a and b if u(x, y) = f (ax + by)

(ii) The function u : R2 → R defined as u(x, t) = (1/2)x2 + t + c, for a given

(iii) Note that u(x, t) = eax+bt is a solution to the PDE ut = uxx if b = a2 .

for given functions f and aα (|α| ≤ k). If f ≡ 0, we say F is homogeneous.

We say F is quasilinear if it has the form

linear ⊂ semilinear ⊂ quasilinear ⊂ fully nonlinear.

Example 1.7. (i) a1 (x)uxx + a2 (x)uxy + a3 (x)uyy + a4 (x)ux + a5 (x)uy =

Schrödinger Equation iut + ∆u = 0.

Korteweg de Vries (KdV) Equation ut + ux + uux + uxxx = 0.

(i) (y − u)ux + xuy = xy + u2 .

(ii) uu2x − xuy = x2 u3 .

(iii) x2 ux + (y − x)uy = y sin u.

(iv) (sin y)ux − ex uy = ey u.

(vi) uux + x2 uyyy + sin x = 0.

(viii) utt + (sin y)uyy − et cos y = 0.

(ix) x2 uxx + ex u = xuxyy .

(x) ey uxxx + ex u = − sin y + 10xuy .

(xi) y 2 uxx + ex uux = 2xuy + u.

(xii) ux uxxy + ex uuy = 5x2 ux .

(xiii) ut = k 2 (uxx + uyy ) + f (x, y, t).

(xiv) x2 uxxy + y 2 uyy − log(1 + y 2 )u = 0.

(xvii) uuxx + uyy − u = 0.

(xviii) uxx + ut = 3u.

Exercise 5. Rewrite the following PDE in the new coordinates v and w.

(i) ux + uy = 1 for v = x + y and w = x − y.

(ii) aut + bux = u for v = ax − bt and w = t/a where a, b 6= 0.

(iii) aux + buy = 0 for v = ax + by and w = bx − ay, where a2 + b2 > 0.

(iv) utt = c2 uxx for v = x + ct and w = x − ct.

(v) uxx + 2uxy + uyy = 0 for v = x and w = x − y.

(vi) uxx − 2uxy + 5uyy = 0 for v = x + y and w = 2x.

(viii) uxx + 2uxy − 3uyy = 0 for v = y − 3x and w = x + y.

1.2 Well-Posedness of PDE

(a) has a solution (existence)

(b) the solution is unique (uniqueness)

(c) the solution depends continuously on the data given (stability).

Example 1.10. The initial value problem

which has the solution uε (x, t) = ε2 sin(x/ε) sin(t/ε). Since

sup{|uε (x, t) − u(x, t)|} = ε2 sup {|sin(x/ε) sin(t/ε)|} = ε2

the problem is well-posed.

which has the solution uε (x, t) = ε2 sin(x/ε) sinh(t/ε). Since

sup{|uεt (x, t) − ut (x, t)|} = ε sup {|sin(x/ε)|} = ε

lim sup{|uε (x, t) − u(x, t)|} = ε lim sup {|sin(x/ε)|}

the problem is ill-posed. Because a small change in initial data leads to a

Exercise 6. Consider the initial value problem

ut + uxx = 0 (x, t) ∈ R × (0, ∞)