Vous êtes sur la page 1sur 720

Introduction to Classical Partial Differential

Equations

Horst R Beyer123 & Maricela Beyer2,3

Matematicas
Para Todos R

Tuxtla Gutirrez, 2017

1
UNAM, Instituto de Matemticas, Mexico City,
2
Centro Mesoamericano de Fsica Terica (MCTP), Tuxtla Gutirrez,
3
en apoyo al Programa para un Avance Global e Integrado de la
Matemtica Mexicana, FORDECyT.
Outline
First-Order
The Wave Equation in Two
Introduction Quasi-Linear Systems Space Dimensions
Inhomogeneous Wave
Conventions of PDE Equations
Linear Symmetric Conservation Laws for
Basic Definitions Hyperbolic Systems Wave Equations
Problems Compressible Eulers Main Examples of Elliptic
Three Main Examples Equations in One Space Equations
of PDE Dimension The Method of Separation of
Hyperbolic Systems Variables
Problems Simple Transformations of the
Single First-Order The Inhomogeneous Wave Laplace Operator
Equation in One Space Strong Maximum Principles
Quasi-Linear PDE Dimension Greens Representation
The Linear Case Single Second-Order
Integral Curves of Vector Fields Formula
The Linear Case PDE The Solution of the
Problems Dirichlet Problem for
Characteristic Surfaces Classification of Single
Second-Order Balls in Rn
Problems
Time Decay of Solutions of Quasi-Linear PDE Heat Equation
PDE (Energy Method) Second-Order Equations With
The Quasi-Linear Case Constant Coefficients Appendix
Incompressible Eulers Main Examples of
Equations in One Space Hyperbolic Equations Index
Dimension The Wave Equation in Three
Problems Space Dimensions References
Introduction

I This course introduces into the field of classical PDE,


I although a few results emerging from a functional analytic
treatment are stated.
I The latter results will not be used or proved here.
Introduction

I The classical theory of partial differential equations (PDE)


I considers solutions of PDE or systems of PDE
I that are continuous with continuous partial derivatives up
to some order,
I i.e., maps whose corresponding component functions are
of class C k for some k N .
Introduction

I These classes of functions turned out to be inadequate for


deciding such questions as well-posedness,
I i.e., the existence and uniqueness of the solution
I as well as its continuous dependence on given data with
respect to some topology.
I The theory of function spaces
I as well as the mathematical field of functional analysis
were created for that purpose.
Introduction

I Why is the theory of classical PDE still important?


I Still, its results provide the basic ways of thinking and the
experience that shape the field.
I Without this intuition more advanced courses in PDE
would be hard to follow.
Introduction

I We enter the field through quasi-linear partial differential


equations of the first order.
I Each of these can be solved in a geometric way:
I The restriction of the unknown function of such a PDE to
so called characteristic curves
I leads to a family of ordinary differential equations.
I The solution of these ODE leads to the solution of the PDE.
I The characteritics can be read off from the PDE in
question.
Introduction

I The solutions of such PDE show a remarkable complexity.


I In particular, the possibility or impossibility of the solution
of a boundary value problem is dependent on the geometry
of the characteristics.
I Such PDE provide an excellent training ground for what to
expect in the solution of higher order equations that will be
addressed later on in the course.
Introduction

I References: There are a couple of good references in the


field of classical PDE, such as [9, 12, 23],
I but these and other approaches I am aware of lack
geometrical viewpoints
I and use somewhat dated notation that affects the
transparency of the results.
I The current approach uses geometrical point of views
wherever possible
I and modern notation throughout to achieve utmost
transparency.
Introduction

I For the following, it should be helpful to keep in mind that


predominantly, the theory of PDE is trying to answer the
following questions:
1. Is there a solution of the problem, usually a PDE or a
systems of PDE subject to boundary and/or initial
conditions.
2. If there is such a solution, is it unique?
Introduction

3. If there is unique solution:


I Is it stable against a small perturbation of the data, e.g., the
boundary and/or initial conditions or even a small
perturbation of the PDE/system of PDE?
I How can the solution be calculated or at least
approximated? In the latter case, what about error bounds?
I If a calculation of the solution is not feasible, are there any
special properties of the solution, e.g., how smooth
(differentiable) is it, or is there an asymptotic behavior?
I
Conventions

I The symbols N, R, C been excluded.


denote the natural I We call x R positive
numbers (including zero),
(negative) if x > 0
all real numbers and all
(x 6 0).
complex numbers,
respectively. I In addition, we call x R
I The symbols N , R , C strictly positive (strictly
denote the corresponding negative) if x > 0
sets from which 0 has (x < 0).
Conventions
I For every n N , e1 , . . . , en denotes the canonical basis of
Rn ,
I i.e, for every i {1, . . . , n} the i-th component of ei has
the value 1,
I whereas all other components of ei vanish.
I For every x = (x1 , . . . , xn ) Rn , |x| denotes the Euclidean
norm of x given by
q
|x| := x12 + + xn2 .

I Also, we define the Euclidean scalar product of vectors


x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) Rn by

x y := x1 y1 + + xn yn .
Conventions
I In addition, we define the open ball U (x) of radius > 0
around x by

U (x) := {y Rn : |y x| < } ,

I the closed ball B (x) of radius > 0 around x by

B (x) := {y Rn : |y x| 6 }

I and the n-sphere Sn (x) of radius > 0 around x by

Sn (x) := {y Rn : |y x| = } .

I In the last symbol, the index is omitted if = 1 and the


label x and the brackets are omitted if x = 0.
I In addition, for every S Rn , S denotes the closure of S in
the Euclidean topology.
Conventions

I Further, in connection with matrices, the elements of Rn


are considered as column vectors.
I In this connection, for K {R, C}, M(n n, K) denotes
the vector space of n n matrices with entries from K.
I For every A M(n n, K),
I det A denotes its determinant
I and
kerA := {x Rn : A x = 0} ,
I where the dot denotes matrix multiplication.
Conventions

I We always assume composition of maps (which includes


addition, multiplication etc.) to be maximally defined.
I For instance, the addition of two maps is defined on the
(possibly empty) intersection of their domains.
I For every non-empty set S, idS denotes the identical map
on S defined by
idS (p) := p
for every p S.
Conventions
For each k N, n N , K {R, C} and each non-empty open
subset of Rn ,
I the symbol C k (, K) denotes the linear space of
continuous and k-times continuously partially
differentiable K-valued functions on .
I Further, C0k (, K) denotes the subspace of C k (, K)
containing those elements that have a compact support in
.
I In addition, if U Rn is non-open and such that
U , then C k (U, K) is defined as the subspace of
C k (, K) consisting of those elements for which there is an
extension to an element of C k (V , K) for some open subset
V of Rn containing U.
I In the special case k = 0, the corresponding superscript is
omitted.
Conventions

For every map f : U Rn defined on some subset U Rm ,


where m, n N , and differentiable in x U,
I f 0 (x) M(m n, R) denotes the derivative of f in x given
by
fi
f 0 (x)ij := (x)
xj
for all i {1, . . . , n} and j {1, . . . , m}.
I In addition, in the case n = 1, we define the gradient of f in
x by
n
X f
(f )(x) := (x) ei .
i=1
x i
Conventions

I Further, for every differentiable map from some


non-trivial open interval I of R into Rn , we define

0 ( ) := ((p1 ) 0 ( ), . . . , (pn ) 0 ( ))

for every I,
I where p1 , . . . , pn denote the coordinate projections of Rn ,
defined by
pi (x) := xi
for every x = (x1 , . . . , xn ) Rn and i {1, . . . , n}.
Basic Definitions
Definition 3.1
I A partial differential equation (PDE) for a map
u : Rp , p N ,
I where is a non-trivial open subset of some Rn for n N
satisfying n > 2,
I is an equation

2u 2u
 
u u
f x, u(x) , (x) , . . . , (x) , 2 (x) , (x) , . . . = 0 ,
x1 xn x1 x1 x2
(3.1)
I where x runs through the elements of .
I Here f : 0 R, where 0 is a non-empty open
subset of some Rm for m N satisfying m > 2p.
Basic Definitions

Definition (cont.)
I Note that such f is not unique.
I The order of the highest derivative occuring in (3.1) 4 is
called the order of the PDE.
I A map u satisfying (3.1) for every x is called a
solution of the PDE.

4
i.e, such that f actually depends on the corresponding coordinate
projection
Basic Definitions

Example 1
(A Hamilton-Jacobi equation)
I The equation
 2
S 1 S
=0 (3.2)
t 2 x
for S : R2 R,
I a special case of the so called Hamilton-Jacobi equation
(for the function S)
I which is of importance in the field of classical mechanics,
I is a partial differential equation.
Basic Definitions

Example (cont.)
I For this, we notice that
 2  
S 1 S S S
(t, x) (t, x) = f t, x, S(t, x) , (t, x) , (t, x)
t 2 x t x

for all (t, x) R2 ,


I where f : R5 R is defined by

x52
f (x) := x4
2
for every x = (x1 , . . . , x5 ) R5 .
Basic Definitions

Example (cont.)
I The order of the highest derivatives occurring in (3.2) is 1,
I and hence the equation is of first order.
I In particular, the family (Sp )pR , where

p2
up (t, x) := t + px
2
for all (t, x) R2 ,
I is a one-parametric family of solutions of the equation.
I The equation is non-linear since u1 is a solution of (3.2),
but 2u1 is not.
Basic Definitions

Example 2
(The wave equation in one space dimension)
I The equation
2u 2u
2 =0, (3.3)
t 2 x
for u : R2 R,
I the so called wave equation in one space dimension (for
the function u)
I which has important applications in the description of the
propagation of waves,
I is a partial differential equation.
Basic Definitions
Example (cont.)
I For this, we notice that

2u 2u
(t, x) (t, x)
t2 x 2
u u
= f t, x, u(t, x) , (t, x) , (t, x) ,
t x
2u 2u 2u 2u

(t, x) , (t, x) , (t, x) , 2 (t, x)
t 2 tx xt x

for all (t, x) R2 ,


I where f : R9 R is defined by

f (x) := x6 x9

for every x = (x1 , . . . , x9 ) R9 .


Basic Definitions

Example (cont.)
I The order of the highest derivatives occurring in (3.3) is 2,
I and hence the equation is of second order.
I Also, as a consequence of the sum rule and product rule for
partial derivatives, the equation is linear.
I In particular, the family (u(a,b) )(a,b)R2 , where

u(a,b) (t, x) := at + bx

for all (t, x) R2 ,


I is a two-parametric family of solutions of the equation.
Basic Definitions

I Examples for PDE equations without solutions are easy to


find.
I For instance, the PDE
 2
u
= 1
x

for partially differentiable u : R2 R has no solution.


I For a less obvious example of system of PDE without
solutions, see Problem 3.
Basic Definitions

Definition 3.2
I A system of partial differential equations for u
I consists of a finite number of partial differential equations
(PDE) for u,
I and u is called a solution of the system if it satisfies all
those equations.
I The order of a system of PDE is defined as the maximum of
the orders of those partial differential equations.
Basic Definitions

Definition 3.3
I A PDE or system of PDE is called linear
I if sums as well as real multiples, both pointwise defined, of
solutions are solutions, too.
I Otherwise, it is called non-linear.
Basic Definitions

Example 3
(Cauchy-Riemann equations)
I The system

u1 u2 u1 u2
= , = ,
x1 x2 x2 x1

for u = (u1 , u2 ) : R2 R2 ,
I the so called Cauchy-Riemann system of differential
equations (for u)
I which has applications in the field of complex analysis,
I is a system of PDE.
Basic Definitions

Example (cont.)
I For this, we notice that
u1 u2
(x) (x)
x1 x2
 
u1 u2 u1 u2
= f1 x, u1 (x), u2 (x), (x) , (x) , (x) , (x) ,
x1 x1 x2 x2
u1 u2
(x) + (x)
x2 x1
 
u1 u2 u1 u2
= f2 x, u1 (x), u2 (x), (x) , (x) , (x) , (x) ,
x1 x1 x2 x2
Basic Definitions

Example (cont.)
I where f1 , f2 : R7 R are defined by

f1 (x1 , . . . , x7 ) := x4 x7 , f2 (x1 , . . . , x7 ) := x5 + x6

for every x = (x1 , . . . , x7 ) R7 .


I The order of the highest order derivatives occurring in the
system is 1,
I and hence the system is of first order.
I Also, as a consequence of the sum rule and product rule for
partial derivatives,
I the system is linear.
Basic Definitions

Example (cont.)
I In particular, the family (u)(a,b)R2 , where

u(a,b) (x) := (a, b)

for all x R7 ,
I is a two-parametric family of solutions of the system.
Basic Definitions

1) For each of the following PDE or systems of PDE, find a


map f such that

2u 2u
 
u u
f x, u(x) , (x) , (x) , 2 (x) , (x) , . . . = 0 ,
x1 x2 x1 x1 x2

where u : Rp and p is an appropriate element of N ,


and x runs through the elements of R2 . In particular, find
the order n of the equation or system and decide whether it
is linear. Also, if the equation or system is linear, find a
pn-parametric family of solutions. On the other hand, if the
equation or system is non-linear, find a non-constant
solution and use the solution for the proof of non-linearity.
Basic Definitions
a)
2u 2u u
2
(t, x) 2
(t, x) = 3 (t, x) , := R2 ,
t x t
b)
2u 2u
(x, y) = (x, y) , := R2 ,
x 2 y2
c)
4u 4u 4u
(x, y) + 2 (x, y) + (x, y) = 0 , := R2 ,
x 4 x 2 y2 y4
d)
2u 2u
2
2 = u4 , := (0, ) R ,
t x
e)
u1 u2 u2 u1
+ =0, + = 0 , := R2 .
t x t x
Basic Definitions

Remark: The equation from a) is a special case of the telegraph


equation in 2-dimensions, the equation from b) is the so called
Laplace equation in 2-dimensions, the equation from c) is the so
called biharmonic equation in 2-dimensions, equations of the
form of that in d) appear in Quantum Field Theory and the
system from e) is the system of Maxwells evolution equations
in 1-space dimension.
Basic Definitions
2) Let U be some non-empty open subset of R2 and
a : U R, b : U R partially differentiable such that5
a a b b
b a =b a =0.
x y x y
Show that u : U R defined by

u(x, y) := f (a(x, y)x + b(x, y)y)

for all (x, y) U and some differentiable f : R R


satisfies
u u
b a =0.
x y

5
Note that the following equations are trivially satisfied if a and b are both
constant functions.
Basic Definitions
3) Let p : R R be differentiable, u a real-valued function
that is defined as well as partially differentiable on a
non-empty open subset U of R2 such that
u u
= (p u) .
t t
a) Show for every differentiable f : R R, that the function
v defined by
v(t, x) := f (x + p(u(t, x)) t)
for every (t, x) U, satisfies
v v
= (p u) .
t t
b) In this way, find non-constant u defined on some
non-empty open subset on R2 satisfying
b1)
u u
=a ,
t x
Basic Definitions
4) Show that u : R defined by
2
u := (3.4)
x
satisfies
u 2u u
2 +u =0. (3.5)
t x x
Here, R2 is non-empty and open, R, t and x
denote the coordinate projections of R2 onto the first and
second coordinate, respectively, and C 3 (, R ) is a
solution of the heat equation in one space dimension

2
2 =0 .
t x
Basic Definitions

Remark: (3.5) coincides with the Navier-Stokes equation


describing the evolution of the velocity distribution u of an
imcompressible fluid in one space dimension. Sometimes, this
equation is also called Burgers equation. (3.4) is called
Hopf-Cole transformation. Note that the latter transforms a
nonlinear equation into a linear equation.
Basic Definitions

5) Let r > 0 and a R. Show that the system of PDEs


u u
(x, y) = ax 2 y + 1 , (x, y) = 4x 3
x y

for every (x, y) Ur (0) R2 and differentiable


u : Ur (0) R satisfying u(0, 0) = 0 has a solution if and
only if a = 12. In the latter case, show that there is a
unique solution such that u(0, 0) = 0. In addition,
calculate this solution.
Basic Definitions

6) a) Let f : R R and g : R R be twice differentiable


functions. Define

u(t, x) := f (x t) + g(x + t)

for all (t, x) R2 . Calculate

u u 2u 2u
(t, x) , (t, x) , (t, x) , (t, x)
t x t 2 x 2
for all (t, x) R2 .
a1) Conclude that u satisfies the wave equation in one space
dimension
2u 2u
2 =0. (3.6)
t 2 x
Basic Definitions
t

u Hb ,a L
b

a u Ha ,b L

x
-b -a O a b

u H-a ,-b L -a

-b
u H-b ,-a L

Figure 1: Visualization of the parallelogram identity, (3.7). The sum


of the values of u in the red points equals that in the blue points.
Basic Definitions

6) a) a2) Prove that the following parallelogram identity is true

u(ta, xb)+u(t+a, x+b) = u(tb, xa)+u(t+b, x+a)


(3.7)
for all (t, x), (a, b) R2 .
b) Find solutions u1 : R2 R, u2 : R2 R of (3.6)
satisfying the following initial conditions

1 u1
u1 (0, x) = , (0, x) = 0 ,
1 + x2 t
u2 1
u2 (0, x) = 0 , (0, x) =
t 1 + x2
for all x R.
Basic Definitions

7) Find all maximal solutions of the minimal surface equation


"  2 # 2 "  2 # 2
u u u u u u 2 u
1+ + 1 + 2 =0
y x 2 x y2 x y xy
p
of the form u(x, y) = k( x 2 + y2 ), (x, y) R2 \ Ba (0),
where k C 2 ((a, ), R) and a > 0.
Basic Definitions

8) (Transformations of partial differential expressions) Let


u : R2 R be twice differentiable.
a) Let A, B, C, a, d R and v := u h, where h : R2 R2 is
defined by
h(t, x) := (at, dx)
for all (t, x) R2 . Show that

2v 2v 2v
A 2 (t, x) + B (t, x) + C 2 (t, x) =
 t 2 tx
2u
x 
u 2u
a2 A 2 + 2adB + d 2 C 2 (h(t, x))
t tx x

for all (t, x) R2 .


Basic Definitions

8) b) Let A, B, C, a, b, c, d R and v := u h, where


h : R2 R2 is defined by

h(t, x) := (at + bx, ct + dx)

for all (t, x) R2 . Show that

2v 2v 2v
A 2 (t, x) + B (t, x) + C 2 (t, x) =
 t tx
2
x
u 2u
(a2 A + 2abB + b2 C) 2 + 2[acA + (ad + bc)B + bdC] +
t tx
2u

(c2 A + 2cdB + d 2 C) 2 (h(t, x))
x

for all (t, x) R2 .


Basic Definitions

8) b) Note that

a2 A + 2abB + b2 C = ht(a, b)|M t(a, b)i ,


c2 A + 2cdB + d 2 C = ht(c, d)|M t(c, d)i ,
acA + (ad + bc)B + bdC = ht(a, b)|M t(c, d)i ,

where M denotes the symmetric 2 2-matrix


 
A B
M := ,
B C
tdenotes transposition, matrix multiplication and h | i the
Euclidean scalar product on R2 .
Basic Definitions

8) c) Use the results of a) and b) to find a, b, c, d R such that


v := u h, where h : R2 R2 is defined by

h(t, x) := (at + bx, ct + dx)

for all (t, x) R2 , satisfies

2v 14 2 v 2v
(t, x) + (t, x) + 5 (t, x)
t 2 3 tx x 2
2u 2u

= 2 (h(t, x))
t 2 x

for all (t, x) R2 .


Three Main Examples of PDE
I For the following examples, let be a non-empty open
subset of R3 .
I Poissons equation determines,
I for example6 , the gravitational potential as a function of
position, U C 2 (R3 , R),
I generated by a mass distribution C(R3 , [0, )).
I In this application, it is given by
4U = 4 ,
I where as usual
3
X 2U
4U :=
i=1
xi2
I and 7 is the gravitational constant.
6
Another application of Poissons equation is in the determination of a
static electric field from a static charge distribution in electrodynamics.
7
= (6.67428 0.00067) 1011 m3 kg1 s2 [5].
Three Main Examples of PDE

I The gravitational field corresponding to U is given by

Eg := U .

I Poissons equation is of second order.


I For the case of a vanishing mass distribution, it is linear.
I It is a prototype of a so called elliptic equation.
Three Main Examples of PDE

In the following examples, I denotes a non-empty open interval


of R whose elements are interpreted as specifying time.
Therefore, the partial derivative of a function f defined on I
into the direction of the first coordinate is denoted by
f
.
t
Three Main Examples of PDE
I The heat equation8 describes the temperature as a function
of position and time, T C 2 (I , R),
I in a homogeneous body occupying the volume .
I It is given by
T
(t, x) [4T (t, )](x) = 0
t
for every (t, x) I ,
I where (0, ) is the thermal diffusivity9 of the body.
I The heat equation is linear and of second order.
I It is a prototype of a so called parabolic equation.

8
This equation is also sometimes referred as diffusion equation.
9
For instance, the thermal diffusivity of iron at 27 degrees Celsius is about
0.227 cm2 /s [15].
Three Main Examples of PDE
I The wave equation describes,
I in particular, the pressure as a function of position and
time, p C 2 (I , [0, )),
I in a homogeneous body occupying the volume .
I In this application, it is given by

2p
(t, x) c2 [4p(t, )](x) = 0
t 2
for every (t, x) I U,
I where c (0, ) is the speed of sound.10
I The wave equation is linear and of second order.
I It is a prototype of a so called hyperbolic equation.

10
For instance, c is about 343.6 m/s in dry air of 20 degrees Celsius [14].
Three Main Examples of PDE

The definition of ellipticity, parabolicity and hyperbolicity of a


PDE or system of PDE will be given later on in the course.
These characterizations will turn out to have major implications
for the type of boundary value problems that can be posed for
the individual types as well as the qualitative behavior of the
solutions.
Three Main Examples of PDE
1) Calculate all necessary derivatives and determine whether u
is a solution of Laplaces equation in two dimensions on D,
2u 2u
(x, y) + (x, y) = 0
x 2 y2
for all (x, y) D. If applicable, a, b R.

a) u(x, y) := a ex cos(y) + b ex sin(y) for (x, y) D := R2 ,


b) u(x, y) := x 3 3xy2 for (x, y) D := R2 ,
c) u(x, y) := (1/2) ln(x 2 + y2 ) for (x, y) D := R2 \ {0} ,
d) u(x, y) := x sin(x + y) + y cos(x + y) for (x, y) D := R2 ,
 
x+y
e) u(x, y) := arctan
1 xy
for (x, y) D := {(x, y) R2 : xy 6= 1} .
Three Main Examples of PDE
2) Calculate all necessary derivatives and determine whether
u is a solution of Laplaces equation in three dimensions on
D,
2u 2u 2u
(x, y, z) + (x, y, z) + (x, y, z) = 0
x 2 y2 z2
for all (x, y, z) D. If applicable, a, b R.

a) u(x, y, z) := exyz for (x, y, z) D := R3 ,


b) u(x, y, z) := a e5x sin(3y) cos(4z) + b e5x cos(3y) cos(4z)
for (x, y, z) D := R3 ,
c) u(x, y, z) := x 3 2xy2 xz2 for (x, y, z) D := R3 ,
p
d) u(x, y, z) := 1/ x 2 + y2 + z2 for (x, y, z) D := R3 \ {0}
Three Main Examples of PDE
3) Calculate all necessary derivatives and determine whether
or, if applicable, when T is a solution of the heat equation
in one space dimension
T 2T
(t, x) 2 (t, x) = 0
t x
for all (t, x) D, where > 0. If applicable, > 0,
a, b R, a1 , . . . , a6 R.
h p  p i
a) T (t, x) := et a sin / x + b cos / x
for (t, x) D := R2 ,

b) T (t, x) := et / x for (t, x) D := R2 ,
c) T (t, x) := a1 + a2 t + a3 x + a4 t 2 + a5 tx + a6 x 2
for (t, x) D := R2 ,
x2
 
1/2
d) T (t, x) := t exp for (t, x) D := (0, ) R
4 t
Three Main Examples of PDE
4) Let f , g : R R be twice differentiable, but otherwise
arbitrary. Define u : R (R2 \ {0}) R by
1
u(t, x) := p [ f ( t |x| ) + g( t + |x| ) ]
|x|
for all (t, x) R (R2 \ {0}). Calculate all necessary
derivatives of u, and verify that u satisfies the wave
equation
 2 
u 1
2
4u (t, x) + u(t, x) = 0
t 4|x|2
for every (t, x) R (R2 \ {0}), where

(4u)(t, x) := [4u(t, )](x)

for all (t, x) R (R2 \ {0}).


Three Main Examples of PDE
5) a) Let f , g : R R be twice differentiable, but otherwise
arbitrary. Define u : R (R3 \ {0}) R by

1
u(t, x) := [ f ( t |x| ) + g( t + |x| ) ]
|x|

for all (t, x) R (R3 \ {0}). Calculate all necessary


derivatives of u, and verify that u satisfies the wave
equation in three space dimensions

2u
4u = 0 ,
t 2
where
(4u)(t, x) := [4u(t, )](x)
for all (t, x) R (R3 \ {0}).
Three Main Examples of PDE

5) b) Consider the case that g = f in part a). Calculate

lim u(t, x)
x0,x6=0

for the corresponding u. Conclude that in general, u cannot


be extended to a twice differentiable function on R2 . An
analogous type of reasoning will be applied in Section 1 in
connection with the so called focusing effect.
Single First-Order Quasi-Linear PDE
I For motivation, we consider a simple example,
I namely the equation
u
(x, y) = g(x, y) (5.1)
y

for every (x, y) R2 ,


I where u : R2 R and g : R2 R is some given
continuous function.
I As a consequence of the definition of partial derivatives,
I the equation (5.1) is equivalent to the family of ordinary
differential equations

(u(x, )) 0 = g(x, ) ,

for x R.
Single First-Order Quasi-Linear PDE
I For x R, it follows by the Fundamental Theorem of
Calculus that
Z y1
u(x, y1 ) u(x, y0 ) = (u(x, )) 0 dy
y
Z 0y1 Z y1
= g(x, ) dy = g(x, y) dy
y0 y0
I and hence, for y0 , y1 R such that y0 6 y1 , that
Z y1
u(x, y1 ) = u(x, y0 ) + g(x, y) dy
y0
11
I We conclude for every y R that
Z y
u(x, y) = u(x, y0 ) + g(x, y) dy .
y0
11
Note that for this to be true, our assumption that g is continuous can be
weakened, for instance, to the assumption that g(x, ) is continuous for every
x R.
Single First-Order Quasi-Linear PDE
y

3
C
1
x
-5 -3 -1 1 3 5
-1

-3

-5

Figure 2: A solution to (5.1) is uniquely determined along a parallel to


the y-axes by its value in one point of that parallel. It is possible to
freely specify the values of such solution only along curves C that
intersect each of those parallels in precisely one point.
Single First-Order Quasi-Linear PDE

I As a consequence, the function u is determined along a


parallel to the y-axes,
I if its value is known in one point of that parallel.
I Therefore, the values of u can be freely specified only on
curves that are intersected at most once by each such
parallel.
I In particular, such values cannot be freely specified on a
curve that is intersected by one of those parallels in two
different points.
Single First-Order Quasi-Linear PDE
I If f : I R,
I where I is some non-empty open interval in R,
I is a given function,
I and the values of u along the graph of f are prescribed,
I the solution to (5.1) on I R,
I assuming those values along that graph,
I is given by
Z y
u(x, y) = u(x, f (x)) + g(x, y) dy
f (x)

for every (x, y) I R.


I We note that, in general, such solution is not necessarily
continuous.
Single First-Order Quasi-Linear PDE

I In the following, also general quasi-linear first-order linear


PDE for one unknown function
I will turn out to be equivalent to families of ordinary
differential equations.
I The role of the parallels to the y-axes is played by the
images of integral curves of a vector field a
I that is read off from the coefficients of the equation.
I For the previous PDE, a is given by a(x, y) = (0, 1) for
(x, y) R2 .
I Such integral curves are considered in the next section.
Single First-Order Quasi-Linear PDE

I In the following, we give a brief introduction to the


geometrical concept of integral curves of vector fields.12
I For this, we consider the special example of the field
a : R2 R2 defined by

a(x, y) := ( 1, 1 + y2 ) (5.2)

for every (x, y) R2 .


I Fig 3 displays the direction field ea : R2 R2 that is
associated to a.

12
A more in depth treatment of this concept is given in courses in
differential geometry or differential topology.
Single First-Order Quasi-Linear PDE

I It is defined by

ea (x, y) := |a(x, y)|1 a(x, y)


!
2
1 1+y
= p , p
2
2 + 2y + y 4 2 + 2y2 + y4

for every (x, y) R2 .


I As a consequence, for every (x, y) R2 , the directions of
a(x, y) and ea (x, y) coincide and, in addition, ea (x, y) has
length one
|ea (x, y)| = 1 .
Single First-Order Quasi-Linear PDE
1.5

1.0

0.5

y 0.0

-0.5

-1.0

-1.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
x

Figure 3: Plot of the direction field ea corresponding to a defined in


(5.2) along with the image of an integral curve of a.
Single First-Order Quasi-Linear PDE
I For motivation, we interpret a as a velocity field,
I e.g., of a plane fluid flow.
I If we expose a point particle of negligible mass to such a
flow,
I the particle will be carried along,
I such that at every time t
I its corresponding speed coincides with the speed of the
fluid flow at its position at that time.
I Hence if : I R2 describes the path of the particle as a
function of time,
I where I is some non-trivial open time interval,

0 (t) = a((t)) ,

for all t I.
Single First-Order Quasi-Linear PDE

I Such path is called an integral curve of a.13


I Note that in this case also : ( c, c) R2 ,
I where , R are such that I = (, ),
I defined by
(t) := (t + c)
for every t ( c, c),
I is an integral curve of a with the same range as .

13
Here, some abuse of terminology is to be noticed. Usually, curves are
defined as particular sets, whereas is a parametrization of such a set,
namely, the image (I) of . Hence a more accurate, but less common,
naming convention for integral curves would be integral paths.
Single First-Order Quasi-Linear PDE
I For this reason, in the following, we can restrict
consideration to integral curves with domains containing 0.
I If j is the j-th component function of , j {1, 2},
I we arrive at the following system of ordinary differential
equations:

10 (t) = 1 , 20 (t) = 1 + (2 (t))2 ,

t I.
I The first equation has the solution

1 (t) = t + x0 ,

t I,
I where x0 := 1 (0), y0 := 2 (0).
Single First-Order Quasi-Linear PDE

I The second equation is equivalent to

20 (t)
1= = (arctan 2 ) 0 (t) ,
1 + (2 (t))2
t I.
I Hence
arctan(2 (t)) = t + arctan(y0 ) ,
t I.
Single First-Order Quasi-Linear PDE
I From this, we conclude that

I Im := ( arctan(y0 ) (/2), arctan(y0 ) + (/2))

I and that
2 (t) = tan(t + arctan(y0 ))
for all t I Im .
I As a consequence, the integral curve satisfying
(0) = (x0 , y0 )
I and with maximal domain is given by

(t) = (t + x0 , tan(t + arctan(y0 ))) ,

t Im .
I Fig 3 displays part of the range of this maximal integral
curve corresponding to x0 = y0 = 0.
Single First-Order Quasi-Linear PDE

Definition 5.1
(Integral Curves)
I Let n N , be a non-empty open subset of Rn

I and a C 1 (, Rn ).

I An integral curve of a is a differentiable path

I from some non-empty open interval I of R into

I satisfying
0 (t) = a((t)) ,
for every t I.
Single First-Order Quasi-Linear PDE

According to theory of ordinary differential equations14 , the


following existence and uniqueness theorem holds
[4, 6, 11, 22].15

14
more precisely, as a consequence of the theorem of Picard and Lindeloef
and by application of the axiom of choice
15
For the validity of Theorem 5.2, it is sufficient that a is locally Lipschitz
continuous. On the other hand, since the course considers only solutions of
PDE systems of class C k for some k N , such generality is not needed in
the following.
Single First-Order Quasi-Linear PDE

Theorem 5.2
(Existence and uniqueness of integral curves) Let n N ,
be a non-empty open subset of Rn , a C 1 (, Rn ), x and
t0 R. Then there is a uniquely determined maximal, i.e.,
inextendable, integral curve such that (t0 ) = x.
Single First-Order Quasi-Linear PDE

As a consequence of the previous theorem, maximal integral


curves with different ranges cannot intersect.
Corollary 5.3
Let n N , be a non-empty open subset of Rn ,
a C 1 (, Rn ) and 1 : I1 , 2 : I2 be maximal, i.e.,
inextendible, integral curves of a for which there are t1 I1 ,
t2 I2 such that 1 (t1 ) = 2 (t2 ). Then

Ran(1 ) = Ran(2 ) . (5.3)


Single First-Order Quasi-Linear PDE
Proof.
I First, I1 , I2 defined by

I1 := {t R : t + t1 I1 } , I2 := {t R : t + t2 I2 }

I are non-empty open intervals of R around 0


I and 1 , 2 defined by

1 (t) := 1 (t + t1 ) , 2 (s) := 2 (s + t2 )

for every t I1 and s I2


I are maximal integral curves of a such that 1 (0) = 2 (0).
I Hence it follows by Theorem 5.2 that 1 = 2 . The latter
implies (5.3).
Single First-Order Quasi-Linear PDE
For the following vector fields and each point in D, find the
maximal16 integral curve meeting that point at parameter time 0.

a) a(x, y) := (y, x) for (x, y) D := R2 ,


b) a(x, y) := (xy, y2 ) for (x, y) D := R2 ,
c) a(x, y) := (cos2 x sin y, 1) for (x, y) D := R2 ,
d) a(x, y) := (1, cos x / cos y) for (x, y) D := (/6, /6)2 ,
e) a(x, y, z) := (1, y, z) for (x, y, z) D := R3 ,
f) a(x, y, z) := (z, z, x) for (x, y, z) D := R3 ,
g) a(x, y, z) := (1, 1, 1) for (x, y, z) D := R3 .

16
i.e., inextendable
Single First-Order Quasi-Linear PDE

I Such equation is of the form


n
X u
ak + bu = , (5.4)
k=0
xk

I where n N , is a non-empty open subset of Rn+1 ,


I a0 , . . . , an C 1 (, R), b, C(, R) are given
I and u : R is the unknown function which is assumed
to be differentiable.
Single First-Order Quasi-Linear PDE
I The solution of (5.4) can be completely reduced to the
solution of a family of ordinary differential equations by
means of the method of characteristics.
I For this, we read off the characteristic vector field

a := (a0 , . . . , an ) C 1 (, Rn+1 )

from the principal part of (5.4).


I The last is the part containing only the highest order
derivatives of the unknown u, i.e.,
n
X u
ak .
k=0
xk
Single First-Order Quasi-Linear PDE
I The characteristic paths, or short the characteristics of
(5.4) are defined as integral curves of a,
I i.e., paths : I from some non-empty open interval
of R
I whose speed at every position coincides with the
corresponding value of a,
I i.e.,
0 ( ) = a(( )) (5.5)
for every I.
I The range of a characteristic path17 is called a
characteristic curve in the following.
I For the following, it will be sufficient to consider only
characteristic paths defined on intervals around 0.
17
i.e., the set {( ) : I},
Single First-Order Quasi-Linear PDE

According to Theorem 5.2 and its corollary, to every x ,


there corresponds a uniquely determined maximal, i.e.,
inextendable, characteristic path satisfying (0) = x, and such
maximal characteristics 1 and 2 satisfying 1 (0) 6= 2 (0)
cannot intersect.
Single First-Order Quasi-Linear PDE

Along a characteristic path : I , it follows by the chain


rule for differentiable functions in several variables and (5.5)
that

(u ) 0 ( ) = (u)(( )) 0 ( ) = (u)(( )) a(( ))


n
X u
= ak (( )) (( ))
k=0
x k

for every I, where the dot denotes the Euclidean scalar


product for Rn+1 .
Single First-Order Quasi-Linear PDE

Hence the restriction of (5.4) to gives


n
X u
( )( ) = (( )) = ak (( )) (( )) + b(( )) u(( ))
k=0
xk
0
= (u ) ( ) + (b )( ) (u )( )

for every I.
Single First-Order Quasi-Linear PDE

This is an ordinary linear differential equation of the first order


for u . Its solution is given by
 Z 
(u )( ) = u(x) + ( )(s) eB (s)
ds eB ( )
0

for every I, where x := (0) and B C 1 (I, R) is the


antiderivative of b satisfying B (0) = 0. Summarizing, we
proved the following theorem.
Single First-Order Quasi-Linear PDE
Theorem 5.4
Let n N , be a non-empty open subset of Rn+1 ,
a0 , . . . , an C 1 (, R), b, C(, R) and u : R be
differentiable such that
n
X u
ak + bu = . (5.6)
k=0
xk

Finally, let : I , where I is some non-empty open interval


around 0, be a characteristic of this equation, i.e., an integral
curve of the characteristic vector field

a := (a0 , . . . , an ) C 1 (, Rn+1 )

of (5.6).
Single First-Order Quasi-Linear PDE

Theorem (cont.)
Then
 Z 
(u )( ) = u(x) + ( )( ) eB ( )
d eB ( ) (5.7)
0

for every I, where x := (0) and B C 1 (I, R) is the


antiderivative of b satisfying B (0) = 0.
Single First-Order Quasi-Linear PDE

Remark 5.4.1
Note in particular that u is constant along characteristics if the
functions b and in (5.6) both vanish.
Remark 5.4.2
Hence the values of u along a characteristic of (5.6) are
uniquely determined by its value in x = (0). As a
consequence, the values of u can only be freely specified on
surfaces that are intersected at most once by the maximal
characteristic curves. In particular, such values cannot be
freely specified on any surface in that is intersected by a
single characteristic curve in two different points.
Single First-Order Quasi-Linear PDE

Figure 4: Characteristic curves and a data surface S. The point p is the


intersection of the characteristic curve through x with S.
Single First-Order Quasi-Linear PDE

We conclude that, the calculation of the value of u at a point


x proceeds in three steps.
1) In the first step, the maximal characteristic path through
x is found such that (0) = x.
2) In the next step, the parameter value is found when that
characteristic intersects a given data surface S, where
values of the solution are prescribed.18

18
In case that does not intersect S, the initial value problem has no unique
solution.
Single First-Order Quasi-Linear PDE

3) The third step uses (5.7) to calculate u(x):


Z
B ( )
u(x) = u((0)) = (u)( ) e ()( ) eB ( ) d ,
0

where B C 1 (I, R) is the antiderivative of b


satisfying B (0) = 0.
4) If in this way a differentiable function u : R is
defined, in the last step, it needs to be verified that u
satisfies (5.6) and assumes the prescribed data on S. If both
is the case, u is the uniquely determined differentiable
solution of (5.6) assuming the prescribed values on S.
Single First-Order Quasi-Linear PDE
2

y 0

-1

-2

-2 -1 0 1 2
x

Figure 5: Plot of the normalized characteristic vector field in


Example 4 along with eight characteristic curves and the unit circle S 1 .
Single First-Order Quasi-Linear PDE

Example 4
I For this, let m R. We consider the equation
u u
x (x, y) + y (x, y) = m u(x, y) (5.8)
x y

for differentiable u : R, where := R2 \ {(0, 0)}.


I The components x , y of a maximal characteristic path
satisfy
x0 ( ) = x ( ) , y0 ( ) = y ( )
for every in its domain
Single First-Order Quasi-Linear PDE

Example (cont.)
I and hence are given by

x ( ) = x (0) e , y ( ) = y (0) e

for every R.
I As a consequence, the range of every maximal
characteristic path is a half-ray originating from the origin
minus the origin.
I The antiderivative of (R R, 7 m) vanishing in 0 is
given by m idR .
Single First-Order Quasi-Linear PDE

Example (cont.)
I Hence it follows by (5.7) that

u(e (0)) = (u )( ) = u((0)) em

for all R.
Single First-Order Quasi-Linear PDE

Example 5
Let f : S 1 R be continuous and u : R be a differentiable
solution of (5.8) with initial values f on S 1 , i.e., such that

u(x, y) = f (x, y) (5.9)

for all (x, y) S 1 . Calculate u.


Single First-Order Quasi-Linear PDE

Example (cont.)
Solution: Using the notation of Example 4, it follows for
(x, y) that

u(x, y) = u (x 2 + y2 )1/2 .(x 2 + y2 )1/2 .(x, y) = u(e .(0))




= u((0)) (e )m ,

where
1
:= ln(x 2 + y2 ) , (0) := (x 2 + y2 )1/2 .(x, y)
2
Single First-Order Quasi-Linear PDE

Example (cont.)
and hence that

u(x, y) = f (x(x 2 +y2 )1/2 , y(x 2 +y2 )1/2 )(x 2 +y2 )m/2 . (5.10)

Indeed, if there is an extension of f to a function f which is


defined as well as differentiable on an open neighborhood of S 1 ,
Single First-Order Quasi-Linear PDE

Example (cont.)
it follows that
u
 f

2 2 (m3)/2 2 f
(x, y) = (x + y ) y (x, y) xy (x, y)
x x y
+ mxf (x, y)(x + y )
2 2 (m2)/2

f
 
u 2 2 (m3)/2 2 f
(x, y) = (x + y ) xy (x, y) + x (x, y)
y x y
+ myf (x, y)(x 2 + y2 )(m2)/2 ,

where
x := x(x 2 + y2 )1/2 , y := y(x 2 + y2 )1/2
Single First-Order Quasi-Linear PDE

Example (cont.)
and hence that the corresponding u satisfies (5.8) and (5.9). In
particular, such u is uniquely determined. Note that u : R
defined by (5.9) is
(i) homogeneous of degree m, i.e., satisfies

u(a.(x, y)) = am u(x, y)

for all a (0, ) and (x, y) ,


(ii) and extendable to a continuous function on R2 for a
non-vanishing f only if m > 0.
Single First-Order Quasi-Linear PDE

Example (cont.)
In addition, note that
(iii) if we consider (5.8) on := R2 which includes the point
where the principal part of (5.8) vanishes19 , the existence
of a solution would depend on m.

19
Such a point is called a singular point of a PDE.
Single First-Order Quasi-Linear PDE

Example (cont.)
(iv) in general, it is not possible to freely specify the boundary
values of u along a closed curve in such as the circle of
radius 1/2 around the point (1, 1). The non-existence of
solutions satisfying boundary values on closed surfaces in
the domain is typical for hyperbolic PDE and systems of
PDE. That such specification is possible for (5.8) and some
closed curves encircling the origin such as S 1 is due to its
singular nature at at the origin which is connected to the
vanishing in the origin of the extension of the
corresponding characteristic vector field to a continuous
function on R2 .
Single First-Order Quasi-Linear PDE

Example (cont.)
In addition, note that (5.10) allows the definition of generalized
solutions for (5.8), i.e., solutions which are not differentiable.
Single First-Order Quasi-Linear PDE

Example 6
(Continuity equation/Conservation of mass)
I Important examples of linear first-order PDE for single
functions are so called continuity equations or
conservation laws. A particular example is the equation
describing the conservation of mass in fluid dynamics
I

+ div(v) = 0 . (5.11)
t
Single First-Order Quasi-Linear PDE

Example (cont.)
I Here : R [0, ) is the mass density of a fluid
occupying the volume ,
I the latter is assumed a non-empty and open subset of R3 ,
I v : R R3 is the distribution of the fluids speed,
I and v is the mass flux density of the fluid.
I In particular, we assume that is differentiable and that
v C 1 (R , R).
Single First-Order Quasi-Linear PDE

Example (cont.)
Equation (5.11) is equivalent to

+ v grad() + div(v) = 0 .
t
The characteristic field of the last equation is given by

a(t, x) = (1, v(t, x))

for every (t, x) R .


Single First-Order Quasi-Linear PDE

Example (cont.)
I In fluid dynamics, differentiable paths : I R3 , where I
is some non-empty open interval of R, such that

0 (t) = v(t, (t))

for every t I are called particle paths.


I They correspond to paths of idealized fluid particles that
are also called fluid elements.
Single First-Order Quasi-Linear PDE

Example (cont.)
I If is such a particle path,
I then the associated : I R4 defined for every t I by

(t) := (t, (t)) ,

I is a characteristic path of (5.11).


I Hence (5.7) gives
 Z t 
(t, (t)) = (t0 , (t0 )) exp [ div(v)](, ( )) d
t0

for every t I.
Single First-Order Quasi-Linear PDE

Example (cont.)
I This implies that, with increasing time, positive values of
div(v) tend to compress
I and that negative values of div(v) tend to expand fluid
elements.
I Hence incompressible fluids generate speed distributions
with vanishing divergence.
I For instance, for a wide range of applications, water can be
considered incompressible.
Single First-Order Quasi-Linear PDE

I Before we proceed to the definition of characteristic and


non-characteristic surfaces,
I we give a final example that demonstrates the importance
of characteristic curves
I also for the formulation of meaningful boundary
conditions for initial-value problems for (5.4).
Single First-Order Quasi-Linear PDE
t

10

x
1 2 3 4 5

Figure 6: Characteristic curves of (5.12) for the ingoing case +.


Note that, as is usual in physics, the first coordinate, which is a time
coordinate, is drawn on the vertical axis.
Single First-Order Quasi-Linear PDE
t

10

x
1 2 3 4 5

Figure 7: Characteristic curves of (5.12) for the outgoing case .


Note that, as is usual in physics, the first coordinate, which is a time
coordinate, is drawn on the vertical axis.
Single First-Order Quasi-Linear PDE

Example 7
(Connection between characteristic curves and meaningful
boundary conditions)
I For this, let q R.

I We consider the equation

 
u u
= +qu (5.12)
t x
I on the open right half-plane := R (0, ) for
differentiable u : R
Single First-Order Quasi-Linear PDE

Example (cont.)
I and with boundary condition

lim u(t, x) = 0 (5.13)


x0

for all t > 0.


I The characteristic vector field a : R2 is given by

a(t, x) := (1, 1)

for every (t, x) .


Single First-Order Quasi-Linear PDE

Example (cont.)
I Let (t0 , x0 ) . Then the maximal characteristic path
such that (0) = (t0 , x0 ) satisfies

0 ( ) = (1, 1)

for every in its domain,


I and hence : I R2 is given by

( ) = ( + t0 , + x0 )

for every I := (, x0 ) and I := (x0 , ),


respectively.
Single First-Order Quasi-Linear PDE

Example (cont.)
I In particular, C 1 (I, R2 ).
I Note that the continuous extension of to I assumes a
point in the boundary of
I and that, naturally, the continuous extensions of 0 to I \ I
coincides with the value (1, 1)
I of the continuous extension of the characteristic vector
field to the boundary of .
Single First-Order Quasi-Linear PDE

Example (cont.)
I In particular, is ingoing, outgoing at the boundary
I in the sense that the projection of the last extension onto
the inner normal to the boundary,
I (0, 1), is < 0 and > 0, respectively.
I Further, the equation (5.12) is equivalent to
 

v=0,
t x

where v(t, x) := eqx u(t, x) for all (t, x) .


Single First-Order Quasi-Linear PDE

Example (cont.)
I Hence
eqx u(t, x) = eq(xt) u(0, x t)
for all (t, x)
I and therefore
u(t, x) = eqt u(0, x t) (5.14)
for all (t, x) .
Single First-Order Quasi-Linear PDE

Example (cont.)
I Hence for the ingoing case, corresponding to the plus
sign in (5.12),
I the boundary condition (5.13) allows only for the trivial
solution
u(t, x) = 0
for all (t, x) and t > 0.
Single First-Order Quasi-Linear PDE
Example (cont.)
I Whereas for the outgoing case, corresponding to the
minus sign in (5.12),
I such solutions are given by

u(t, x) = eqt f (x t) , (t, x) , (5.15)

I where (
f (x) := 0 for x 6 0
f (x) for x > 0
I and f is some differentiable function on (0, ) satisfying

lim f (x) = lim f 0 (x) = 0 .


x0 x0
Single First-Order Quasi-Linear PDE

The following example demonstrates that, in general, the


asymptotic behaviour of solutions of a PDE is not only
governed by its principal part.
Single First-Order Quasi-Linear PDE
Example 8
(Asymptotic behaviour of solutions)
I For this let q R, we consider the equation

u u
= +qu
t x
on := R2
I for differentiable u : R and initial conditions

u(0, x) = f (x) ,

x R,
I where f : R R is differentiable.
Single First-Order Quasi-Linear PDE

Example (cont.)
I Using the characteristic method, the solution of this initial
value problem is found to be

u(t, x) = eqt f (x + t)

for (t, x) R2 .
Single First-Order Quasi-Linear PDE
Find the differentiable function u that satisfies the given conditions. If
applicable, f : R R and g : R2 R are differentiable.

u u
a) y (x, y) + x (x, y) = 0 , (x, y) R2 \((, 0] {0}) ,
x y
u(x, 0) = x for x > 0 ,
u u
b) xy (x, y) y2 (x, y) + 2(x + y) = 0 , (x, y) R (0, ) ,
x y
u(x, 1) = f (x) for x R ,
 
u 1 u 5y
c) sin(y) (x, y) + (x, y) + 3ye u(x, y) = 0 ,
x cos2 (x) y
(x, y) (/2, /2) R , u(x, 0) = f (x) for x (/2, /2) ,
Single First-Order Quasi-Linear PDE
   
u u
d) cos(y) (x, y) + 2u + cos(x) (x, y) cos(y) = 0 ,
x y
(x, y) (/6, /6)2 , u(0, y) = f (y) for y (/6, /6) ,
u u u
e) (t, x, y) + x (t, x, y) + y (t, x, y) = 0 , (t, x, y) R3 ,
t x y
u(0, x, y) = sin (x + y) for (x, y) R2 ,
2

u u u
f) z (x, y, z) z (x, y, z) x (x, y, z) = 2y , (x, y, z) (0, ) R2 ,
x y z
u(x, y, 0) = g(x, y) for (x, y) (0, ) R ,
u u u
g) (x, y, z) + (x, y, z) + (x, y, z) 3u(x, y, z) = ex+y ,
x y z
(x, y, z) R3 , u(x, y, 0) = g(x, y) for (x, y) R2 .
Single First-Order Quasi-Linear PDE
I We have seen that, in general, the explicit knowledge of the
characteristics of (5.4),
n
X u
ak + bu = , (5.16)
k=0
xk

is necessary
I for the task of deciding whether a boundary value problem
has a solution.
I In general, it is not possible to obtain such knowledge.
I The following will lead to a simpler necessary criterion for
that task.
I This will lead on the concept of characteristic and
non-characteristic surfaces
I which will also turn out important later for second order
PDE.
Single First-Order Quasi-Linear PDE

I If a0 (x) 6= 0 for all x ,


I we can solve (5.16) for the partial derivative of u
I in the direction of the zeroth coordinate:
n
u X ak u b
= u+ . (5.17)
x0 k=1
a 0 xk a0 a 0

I Assuming for the moment that a0 , . . . , an , b, as well as u


are all C ,
I from (5.17), we can compute all partial derivatives of u in
the direction of the zeroth coordinate by differentiation.
Single First-Order Quasi-Linear PDE

For instance,
2u
x02
n n
X (ak /a0 ) u X ak 2 u (b/a0 ) b u (/a0 )
= u +
x0 xk a0 x0 xk x0 a0 x0 x0
k=1 k=1
n n
X (ak /a0 ) u X ak 2 u (b/a0 ) b u (/a0 )
= u + .
x0 xk a0 xk x0 x0 a0 x0 x0
k=1 k=1
Single First-Order Quasi-Linear PDE

Figure 8: Possible data surface x0 = 1 for (5.17) in the case that


n = 2. Note that the first coordinate, which is a time coordinate, is
drawn on the vertical axis.
Single First-Order Quasi-Linear PDE

I Hence, we expect that there is a unique solution to (5.17)


I that coincides with a given C -function on some surface
x0 = constant
I as is displayed by the following example.
I Note the partial derivative in the direction of x0 at a point
on such a surface
I is a partial derivative in a direction which is orthogonal,
I in the Euclidean sense, to the surface in that point.
Example 9
I We consider the PDE
u u
= , (5.18)
t x
I where = R (a, b) for some a, b R such that a < b
I and u : R is assumed infinitely often differentiable.
I We note that (5.18) implies that
nu nu
= , (5.19)
t n x n
for every n N .
Single First-Order Quasi-Linear PDE

Example (cont.)
I For u, we make the following ansatz
N
X 1
u(t, x) = an (x) t n ,
n=0
n!

for all t R and x (a, b),


I where N N and a0 , a1 , . . . , aN C ((a, b), R).
Single First-Order Quasi-Linear PDE

Example (cont.)
I Using (5.19), it follows that
nu nu (n)
an (x) = n
(0, x) = n
(0, x) = a0 (x)
t x
for every n {1, . . . , N}.
I Hence
N
X 1 (n)
u(t, x) = a0 (x) t n ,
n=0
n!
for all t R and x (a, b).
Single First-Order Quasi-Linear PDE

Example (cont.)
I Substitution of the latter into (5.18) leads to the equation
N N1
u X 1 (n)
X 1 (n+1)
(t, x) = a0 (x) t n1 = a0 (x) t n
t n=1
(n 1)! n=0
n!
N
u X 1 (n+1)
= (t, x) = a0 (x) t n
x n=0
n!

I for (t, x) which is satisfied if and only if


(N+1)
a0 =0.
Single First-Order Quasi-Linear PDE

Example (cont.)
I Hence if p : (a, b) R is a polynomial of an order that is
less or equal to N,
I then
N
X 1 (n)
u(t, x) := p (x) t n ,
n=0
n!
for all t R, x (a, b),
I where p(0) := p, is the unique solution of (5.18) in our set
of ansatz functions that satisfies u(0, ) = p.
Single First-Order Quasi-Linear PDE
Example (cont.)
I We note that u defines a polynomial on R2 which is a
particular case of a Taylor series in two variables.
I Indeed, the use of Taylor series in several variables lead to
the first general theorem on the well-posedness of initial
value problems for PDE,
I the Cauchy-Kowalevski theorem.
I Unfortunately, this theorem is usually too narrow for
applications
I since considering only real-analytic solutions
corresponding to real-analytic data.
I For a discussion of this theorem, see, e.g., [12].
Single First-Order Quasi-Linear PDE

That the above expectation is not always correct is


demonstrated by the following two examples, where one of the
coefficient functions is unbounded.
Single First-Order Quasi-Linear PDE
1.0

0.8

0.6

0.4

0.2

0.0

0.0 0.2 0.4 0.6 0.8 1.0


x

Figure 9: Plot of the normalized characteristic vector field in


Example 10 and the curve {(x 2 , x) : x [0, 1]}. Note that the first
coordinate, which is a time coordinate, is drawn on the vertical axis.
Single First-Order Quasi-Linear PDE

Example 10
(Nonuniqueness of solutions, I)
I We consider the PDE

u 1 u
(t, x) = (t, x) , (5.20)
t 2x x
for every (t, x) ,
I where = R (0, ) and u : R is assumed
differentiable.
Single First-Order Quasi-Linear PDE

Example (cont.)
I The maximal characteristic path meeting (t, x) at
parameter time 0 is given by

( ) = ( + t, + x 2 )

for every (x 2 , ).
I This path meets the initial surface {0} (0, ) if and only
if t < x 2 .
I Hence this surface is no suitable data surface.
Single First-Order Quasi-Linear PDE

Example (cont.)
I We note that for every differentiable f : R R,
I the corresponding uf : R defined by

uf (t, x) := f (t x 2 )

for every (t, x) is a solution of (5.20).


I In particular, if f is such that f (x) = 0 for every
x (, 0],
I then uf (0, x) = 0 for every x > 0.
I Hence there is no unique solution of (5.20) corresponding
to differentiable data on (0, ).
Single First-Order Quasi-Linear PDE

Example (cont.)
I For instance, functions f of this type are given by
fa : R R defined by
(
0 if x 6= 0
fa (x) :=
ax 2 if x > 0 .

for every x R, where a R.


Single First-Order Quasi-Linear PDE

I A visual inspection of the normalized characteristic vector


field in Example 10 suggests that,
I in addition to data given on a surface {t0 } (0, ), t0 R,
I also data on the boundary of is needed
I for a unique characterization of a solution to (5.20).
I The following gives an example where such additional data
is needed at spacial infinity.
Single First-Order Quasi-Linear PDE
3

t 0

-1

-2

-3
-3 -2 -1 0 1 2 3
x

Figure 10: Plot of the normalized characteristic vector field in


Example 150 and the curve {(1/x 2 , x) : x [3, 3] \ {0}}. Note that
the first coordinate, which is a time coordinate, is drawn on the
vertical axis.
Single First-Order Quasi-Linear PDE
Example 11
(Nonuniqueness of solutions, II)
I We consider the PDE

u x 3 u
(t, x) (t, x) = 0 , (5.21)
t 2 x
for every (t, x) R2 and differentiable u : R2 R.
I The maximal characteristic path meeting (t, x) at
parameter time 0 is given by

( ) = ( + t, x/ 1 + x 2 )

for every R if x = 0
I and (1/x 2 , ) if x 6= 0.
Single First-Order Quasi-Linear PDE

Example (cont.)
I For the cases that x 6= 0, this path meets the initial surface
{0} R if and only if t < 1/x 2 .
I Hence this surface is no suitable data surface.
I We note that for every differentiable f : R R with
compact support,
I the corresponding uf : R defined by

uf (t, x) := f (t (1/x 2 ))

for every (t, x) R2 is a solution of (5.21).


Single First-Order Quasi-Linear PDE

Example (cont.)
I In particular, if f is such that f (x) = 0 for every
x (, 0],
I then uf (0, x) = 0 for every x R.
I Hence there is no unique solution of (5.21) corresponding
to differentiable data on R.
I We note that, if u is a solution of (5.21),
I then v : R (0, ) R defined by v(t, x) := u(t, 1/x)
for (t, x) R (0, ) is a solution of (5.20).
Single First-Order Quasi-Linear PDE

Example (cont.)
I This connection between the solutions of (5.20) and (5.21)
suggests that,
I in addition to data given on a surface {t0 } R, t0 R,
I also data at spacial infinity is needed for a unique
characterization of a solution to (5.21).
I However, the following theorem confirms that the above
expectation is true under certain assumptions on the
coefficients.
Single First-Order Quasi-Linear PDE

Theorem 5.5
I Let n R , a1 , . . . , an C 1 (Rn , R) which together with all
their partial derivatives are bounded functions.
I In addition, let b C(Rn , R) be bounded and t0 R.
I Then to every u0 W 1 (Rn ) there corresponds a uniquely
determined u : R W 1 (Rn ) satisfying
n
!
X
u 0 (t) = ak u(t) b u(t)
k=1
xk

for all t R and


u(t0 ) = u0 .
Single First-Order Quasi-Linear PDE

Theorem (cont.)
I Here W 1 (Rn ) is the Sobolev space consisting of all
elements of L 2 (Rn ), the space of square integrable
functions on Rn , that posses square integrable weak first
derivatives.
I The derivatives /x1 , . . . , /xk denote weak derivatives
I and 0 denotes the ordinary derivative for L 2 (Rn )-valued
maps.
Single First-Order Quasi-Linear PDE

Proof.
The proof of the statement uses more advanced methods from
functional analysis and will not be given here. The statement
follows, for instance, from Theorem 6.30, Corollary 6.31 in [3]
for autonomous Hermitian hyperbolic systems. For similar
statements for non-autonomous Hermitian and quasi-linear
hyperbolic systems see Theorem 11.8, Remark 11.9 and
Theorem 13.6 in [3].
Single First-Order Quasi-Linear PDE

Figure 11: Sketch of a regular hypersurface S with values (x) of the


normal field and characteristic vector field a(x) in x S.
Single First-Order Quasi-Linear PDE

I The surfaces x0 = constant are special cases of regular


hypersurfaces.
I In the following, we consider more general surfaces of the
last type.
I For this, let f C 1 (, R) such that (x) := (f )(x) 6= 0
for all x S, where

S := {x : f (x) = 0} .

I As a consequence, S is a regular C 1 -hypersurface


contained in with normal field .
Single First-Order Quasi-Linear PDE

I It follows for differentiable u : R and x S that


n
X u
ak = a(x) (u)(x)
k=0
xk
 
a(x) (x) a(x) (x)
= a(x) (x) + (x) (u)(x)
|(x)|2 |(x)|2
 
a(x) (x)
= (x) (u)(x)
|(x)|2
 
a(x) (x)
+ a(x) (x) (u)(x)
|(x)|2

I Here the dot denotes the Euclidean scalar product for Rn+1 .
Single First-Order Quasi-Linear PDE

I Note that
 
a(x) (x)
a(x) (x) (u)(x)
|(x)|2

is the derivative of u into a direction which is normal to


(x) and hence tangential to S in x.
Single First-Order Quasi-Linear PDE
I As a consequence, we can solve (5.4) everywhere on S for
the derivative of u into the normal direction of S
I if and only if a is nowhere vanishing,
I i.e., if and only if is nowhere orthogonal to the
characteristic vector field a.
I If the last is the case, we say that S is non-characteristic.20
I In particular, if S is non-characteristic, it is nowhere
tangent to any characteristic curve.
I In this case, by help of (5.4), the derivative of u in every
point of S into the normal direction of S can be calculated
from the knowledge of the values of u on S.

20
For instance, x0 = c for some constant c R is such surface if a0 (x) 6= 0
for all x .
Single First-Order Quasi-Linear PDE

I On the other hand, we say that S is characteristic if a is


everywhere vanishing,
I i.e., if is everywhere orthogonal to the characteristic
vector field a.
I In this case a is tangential to S, and S contains
characteristic curves.
I Also, by help of (5.4) and for every point in S,
I it is impossible to calculate the derivative of u into the
normal direction of S from the knowledge of the values of
u on S.
Single First-Order Quasi-Linear PDE

As a consequence, non-characteristic regular C 1 -hypersurfaces


are possible candidates for the specification of free data for the
solutions of (5.4), whereas data cannot be freely specified for
these solutions on characteristic surfaces.
Single First-Order Quasi-Linear PDE
Definition 5.6
(Characteristic and non-characteristic hypersurfaces) Let
f C 1 (, R) such that (x) := (f )(x) 6= 0 for all x S,
where
S := {x : f (x) = 0} .
We say that the hypersurface S is non-characteristic if

(x) a(x) = (f )(x) a(x) 6= 0

for all x S, and we say that S is characteristic if

(x) a(x) = 0

for all x S. Note that the latter implies that a|S is tangential to
S and hence that S contains characteristic curves.
Single First-Order Quasi-Linear PDE
For each of the following differential equations, decide whether
the given surfaces are characteristic or non-characteristic.
u u
a) y (x, y) + x (x, y) = 0 , (x, y) R2 ,
x y
1
S1 := S , S2 := R {0} , S3 := (0, ) {0} ,
u u
b) xy (x, y) y2 (x, y) + 2(x + y) = 0 , (x, y) R2 ,
x y
1
S1 := S , S2 := {(x, y) R2 : xy = 1} , S3 := R {1} ,
 
u 1 u 5y
c) sin(y) (x, y) + (x, y) + 3ye u(x, y) = 0 ,
x cos2 (x) y
(x, y) (/2, /2) R , S1 := (/2, /2) {0} ,
S2 := {(x, y) (/2, /2) R : tan(x) + cos(y) = 1} ,
S3 := {0} R ,
Single First-Order Quasi-Linear PDE

   
u u
d) cos(y) (x, y) + 2u + cos(x) (x, y) cos(y) = 0 ,
x y
(x, y) R2 , S1 := (/2, /2) {0} , S2 := R {/2} ,
S3 := {(x, y) R2 : sin(y) sin(x) = 1/2} ,
u u u
e) (t, x, y) + x (t, x, y) + y (t, x, y) = 0 , (t, x, y) R3 ,
t x y
2 2
S1 := {0} R , S2 := R {0} ,
S3 := {(t, x, y) R3 : x + y = et } ,
Single First-Order Quasi-Linear PDE

u u u
f) z (x, y, z) z (x, y, z) x (x, y, z) = 2y ,
x y z
3 2
(x, y, z) R , S1 := {0} R , S2 := (0, ) R {0} ,
u u u
g) (x, y, z) + (x, y, z) + (x, y, z) 3u(x, y, z) = ex+y ,
x y z
(x, y, z) R , S1 := {(x, y, z) R3 : x + y = z} ,
3

S2 := R2 {0} , S3 := {(x, y, z) R3 : x + y 2z = 0} .
Single First-Order Quasi-Linear PDE

In the following, we describe the energy method which leads to


estimates on the asymptotic behaviour of the solutions of PDE
in terms of one of its variables. The basis of the method is
provided by the following lemma.
Single First-Order Quasi-Linear PDE

Figure 12: Sketch of the region [0, T ] S in Lemma 5.7 for the case
S = U1 (0) R2 .
Single First-Order Quasi-Linear PDE
Lemma 5.7
Let n {1, 2, 3}, be an open subset of Rn+1 , T > 0 and S a
nonempty bounded open subset of Rn to which Gauss theorem
is applicable and such that [0, T ] S . Further, let
q C 1 (, Rn+1 ) and , > 0 such that
(i) (div q)(t, x) 6 q0 (t, x)/(t + ) for all (t, x) (0, T ) S,
(ii) (q1 (t, x), . . . , qn (t, x)) (x) > 0 for all
(t, x) (0, T ) S, where : S Rn is the outer unit
normal field of S.
Then
Z   Z
n t
q0 (t, )dv 6 1 + q0 (0, )dvn (5.22)
S S

for all t [0, T ].


Single First-Order Quasi-Linear PDE

Proof.
I For this, we define E : [0, T ] R by
Z
E(t) := q0 (t, )dvn
S

for every t [0, T ].


I By integration of the equation
q0
(t, ) + div(q1 (t, ), . . . , qn (t, )) (div q)(t, ) = 0
t
over S and subsequent application of Gauss theorem,
Single First-Order Quasi-Linear PDE

Proof (cont.).
I we conclude that
Z Z
q0
0= (t, )dv + div(q1 (t, ), . . . , qn (t, ))dvn
n
SZ t S

(div q)(t, )dvn


S Z
0
> E (t) (div q)(t, )dvn
S

for every t (0, T ).


Single First-Order Quasi-Linear PDE

Proof (cont.).
I Hence it follows by application of the fundamental
theorem of calculus that
Z t Z 
n
E(t) 6 E(0) + (div q)(s, )dv ds
0 S
Z t Z 
q0 (s, ) n
6 E(0) + dv ds
0 S s+
Z t
E(s)
= E(0) + ds
0 s+

for every t (0, T ).


Single First-Order Quasi-Linear PDE

Proof (cont.).
I By introduction of the auxiliary function G : [0, T ] R
defined by Z t
E(s)
G(t) := ds
0 s+

for every t [0, T ],


I it follows that
0
(t + )+1 (idR + ) G (t) = (t + )G0 (t) G(t) 6 E(0)


(5.23)
Single First-Order Quasi-Linear PDE

Proof (cont.).
I and hence that
0
(idR + ) G (t) 6 E(0)(t + )(+1)


for every t (0, T ).


I The latter implies that
 
1 1
(t + ) G(t) 6 E(0) (t + ) +

Single First-Order Quasi-Linear PDE

Proof (cont.).
I and hence that
E(0) 
(t + ) 1

G(t) 6

for all t [0, T ].
I Finally, from (5.23), we conclude that

E(t) 6 E(0) + E(0) (t + ) 1 = (t + ) E(0)


 

for every t [0, T ].


Single First-Order Quasi-Linear PDE
Further, we arrive at the following corollary that can be used to
show time decay of quantities that are related to q0 .
Corollary 5.8
In addition to the assumption of Lemma 5.7, let q0 be positive,
U a nonempty open subset of S and a C(U, R) such that

q0 (t, x) > h(t)a(t, x)

for (t, x) [0, T ] U, where h : [0, T ] (0, ). Then (5.22)


implies that

(t + )
Z Z
n
a(t, )dv 6 q0 (0, )dvn
U h(t) S

for all t [0, T ].


Single First-Order Quasi-Linear PDE

Corollary (cont.)
Further, if in addition,
Z Z
n
q0 (0, )dv 6 C a0 (0, )dvn
S U

for some C > 0, then


(t + )
Z Z Z
n 1 n
a(t, )dv 6 q0 (t, )dv 6 C a(0, )dvn
U h(t) S h(t) U

for all t [0, T ].


Single First-Order Quasi-Linear PDE

Remark 5.8.1
I Usually, the previous lemma is applied to a PDE by
multiplication of the equation with a suitable function
(multiplier) of the unknown function
I such that the principal part of the resulting equation is
contained in a term of the form div q
I for some vector field q whose zeroth (time) component is
strictly positive.
Single First-Order Quasi-Linear PDE

Remark (cont.)
I In such a case, the zeroth component is called an energy
density,
I although in general not related to a physical energy
associated with the unknown function,
I and (5.22) can be used to show uniqueness of the solution
to the initial value problem for the original PDE
I and data that are prescribed on surfaces of constant time
(see, e.g., Theorem 5.9).
Single First-Order Quasi-Linear PDE

Remark (cont.)
I Further, Corollary 5.8 can be used to show decay in time of
functions of the unknown function (see, e.g., Example 12).
I Note that in this the existence of a solution of the original
equation is assumed.
I For this reason, in this case, the estimates provided by
Lemma 5.7 and Corollary 5.8 are called a priori estimates.
I Such estimates will also play an important in the study of
elliptic partial differential equations (see Theorem 7.43).
Single First-Order Quasi-Linear PDE

The application of the energy method will be exemplified in the


following theorem and a subsequent example. It is important
to note that the energy method is applicable to non-linear PDEs
including PDEs of higher oder than 1.

The statement of the following Theorem 5.9 is mainly a


consequence of Lemma 5.7.
Single First-Order Quasi-Linear PDE

Theorem 5.9
Let n {1, 2, 3}, T > 0, S a nonempty bounded open subset of
Rn to which Gauss theorem is applicable and be an open
subset of Rn+1 containing [0, T ] S. In addition, let
a1 , . . . , an C 1 (, R), b, C(, R) such that

[div a(t, )](x) 6 2b(t, x) (5.24)

for every (t, x) (0, T ) S, where


a(t, x) := (a1 (t, x), . . . , an (t, x)) for all (t, x) . Further, let
u, v C 1 (, R) such that
Single First-Order Quasi-Linear PDE
Theorem (cont.)

n n
u X u v X v
+ ak + bu = + ak + bv = ,
x0 k=1 xk x0 k=1 xk

at least one of the conditions

a(t, x) (x) > 0

or
u(t, x) = v(t, x) = 0
is satisfied for all (t, x) of (0, T ) S, where : S Rn is
the outer unit normal field of S, and u|{0}S = v|{0}S . Then
u|[0,T ]S = v|[0,T ]S .
Single First-Order Quasi-Linear PDE

Proof.
We define w := u v. Then
n
w X w
= ak bw
x0 k=1
xk

and hence
n
w2 X w2
= ak 2b w2
x0 k=1
xk
n n
!
X ak w 2 X ak
= + 2b w2 .
k=1
xk k=1
xk
Single First-Order Quasi-Linear PDE

Proof (cont.).
As a consequence,
(div q)(t, x) 6 0
for all (t, x) (0, T ) S, where

q := (w2 , a1 w2 , . . . , an w2 ) .

Hence according to Lemma 5.7


Z
w2 (t, )dvn 6 0
S

for all t [0, T ] which implies that w(t, x) = 0 for all


(t, x) [0, T ] S.
Single First-Order Quasi-Linear PDE

Remark 5.9.1
Note that if u : R satisfies
n
u X u
+ ak + bu = ,
x0 k=1 xk

then v : R, defined by v(t, x) := et u(t, x) for every


(t, x) , satisfies
n
v X v
+ ak + (b + ) v = .
x0 k=1 xk
Single First-Order Quasi-Linear PDE

Remark (cont.)
Hence the the statement of Theorem 5.9 is still true if (5.24) is
replaced by the condition that the function

((0, T ) S R, (t, x) 7 [div a(t, )](x) 2b(t, x))

is bounded from above.

Note that Theorem 5.9 applies to both, the ingoing and


outgoing, cases considered in Example 7. The following
example applies Corollary 5.8.
Single First-Order Quasi-Linear PDE

Example 12
We consider the equation
u u
+ =0
t x
on the open right half-plane := R (0, ) for continuously
differentiable u : R and with boundary condition

lim u(t, x) = 0
x0

for all t > 0.


Single First-Order Quasi-Linear PDE

Example (cont.)
Solutions are given by

u(t, x) = f (x t)

for (t, x) , where


(
f (x) := 0 for x 6 0
f (x) for x > 0

and f is some continuously differentiable function on (0, )


satisfying
lim f (x) = lim f 0 (x) = 0 .
x0 x0
Single First-Order Quasi-Linear PDE

Example (cont.)
Note that for every x > 0, all these solutions vanish in points
(t, x) for which t > x. Also note that for every such f , the
corresponding u : R2 R defined by u(t, x) = f (x t) for
every (t, x) R2 is a continuously differentiable function
satisfying
u u
+ =0 (5.25)
t x
and u(0, x) = f (x) for x R.
Single First-Order Quasi-Linear PDE

Example (cont.)
In the following, we try to use Corollary 5.8 to describe this
decay of the solutions for large times, but without the use of the
explicit knowledge of the solutions. For this, let n N . Then
   
2n u u q0 q1
0 = 2(t x) u(t, x) + (t, x) = + (t, x)
t x t x
(5.26)
for every (t, x) R2 , where

q0 (t, x) := q1 (t, x) := (t x)2n u2 (t, x)

for every (t, x) R2 .


Single First-Order Quasi-Linear PDE

Example (cont.)
Hence it follows by Lemma 5.8 for every a > 0, > 0, > 0,
and t > a that
Z a Z a
2n 2
(t a) u (t, x)dx 6 (t x)2n u2 (t, x)dx
0 0
Z a Z a
2n 2 2n
6 (t + ) x u (0, x)dx 6 (t + ) a u2 (0, x)dx
0 0

and hence that


Z a 2n Z a
t

2
u (t, x)dx 6 (t + ) 1 u2 (0, x)dx .
0 a 0
Single First-Order Quasi-Linear PDE

Example (cont.)
This implies that Z a
u2 (t, x)dx = 0
0

for t > a and hence that u(t, x) = 0 for x (0, a] and t > a.
Finally, we conclude that u(t, x) = 0 for x (0, ) and t > x.

The reader might wonder what is the motivation behind the


mutiplication of the equation (5.25) by the multiplicator
(R2 R, (t, x) 7 2(t x)2n (u)(t, x)) in the derivation of the
conservation law (5.26) of Example 12.
Single First-Order Quasi-Linear PDE

Such equation is of the form


n
X u
ak (x, u(x)) (x) = (x, u(x)) (5.27)
k=0
xk

for every x . Here n N , is a non-empty open subset of


Rn+1 , a0 , . . . , an C 1 ( R, R), C 1 ( R, R) are given
and u : R is the unknown function which is assumed to be
C 1 . Note that the quasi-linear generalization of the term bu
from the linear case, (5.4), is absorbed in in the term (x, u(x)).
Single First-Order Quasi-Linear PDE

I Again, by means of the method of characteristics, the


solution of (5.27) can be completely reduced to the
solution of a system of ordinary differential equations.
I The main difference to the case of the linear equation is
that, in general, the characteristic vector field and the
characteristic paths depend on the solution.
I Again, we read off the characteristic vector field

au := (a0 (, u), . . . , an (, u)) C 1 (, Rn+1 )

from the principal part of (5.27)


Single First-Order Quasi-Linear PDE
I which is the part containing the highest order derivatives of
the unknown u, i.e.,
n
X u
ak (, u) .
k=0
xk

I The characteristic paths of (5.27) are defined as integral


curves of au ,
I i.e., maps : I from some non-empty open interval I
around 021 of R whose speed at every position coincides
with the corresponding value of au ,
I i.e.,
0 ( ) = a(( ), (u )( )) (5.28)
for all I.
21
The assumption that 0 I does not restrict generality in the following
since obtainable from any integral curve by a suitable reparametrization.
Single First-Order Quasi-Linear PDE

I Again, the range of a characteristic path22 will be called a


characteristic curve in the following.
I It follows by the chain rule for differentiable functions in
several variables and (5.28) that

(u ) 0 ( ) = (u)(( )) 0 ( )
= (u)(( )) a(( ), (u )( ))
n
X u
= ak (( ), (u )( )) (( )) ,
k=0
x k

where the dot denotes the Euclidean scalar product for


Rn+1 .

22
i.e., the set {( ) : I}
Single First-Order Quasi-Linear PDE

I Hence the restriction of (5.27) to gives


n
X u
(( ), (u )( )) = ak (( ), (u )( )) (( ))
k=0
xk
= (u ) 0 ( )

for every I.
I This is an ordinary differential equation of the first order
for u .
I Differently to the case of the linear PDE, in general, this
equation is in general non-linear.
Single First-Order Quasi-Linear PDE
I Hence we arrive at the following system of ordinary
differential equations for (, u ):

0 ( ) = a(( ), u ( )) , u0 ( ) = (( ), u ( )) (5.29)

for every I, where u := u .


I According to the theory of ordinary differential
equations23 , [4, 6, 11, 22], for every x , there is a
unique maximal, i.e., inextendable, solution to this system
satisfying
(, u )(0) = (x, ux ) ,
where (x, ux ) R.
I Note that still the values of u along are uniquely
determined by its value in x = (0).
23
more precisely, as a consequence of the theorem of Picard and Lindeloef
and by application of the axiom of choice
Single First-Order Quasi-Linear PDE

I As a consequence, the values of u can be freely specified


only on surfaces that are intersected at most once by the
maximal characteristic paths.
I In particular, such values cannot be freely specified on any
surface in that is intersected by a single characteristic
path in two different points.
I We also note that if (, u ) is a solution of (5.29),
I also (, u )( + c) is a solution of (5.29) for every c R.
I As a consequence of (5.29), in general, the characteristic
paths of (5.27) depend on the solution u.
Single First-Order Quasi-Linear PDE

I On the other hand, in the case that the coefficient function


a in that equation does not depend on u,
I i.e., if a(x, y) = a(x, y 0 ) for all (x, y), (x, y 0 ) R,
I those characteristic paths are not dependent on u.
I In the latter case, the equation (5.27) is called semi-linear.
I In that case, the calculation of the value of u at a point
x , typically, proceeds in three steps that are similar to
that of the linear case.
Single First-Order Quasi-Linear PDE

I Steps in the solution of the semi-linear case:


1) In the first step, the maximal characteristic path through x
is found such that (0) = x.
2) In the next step, the parameter value is found such that
the corresponding ( ) is in a given data surface S, where
values of the solution are prescribed.
3) In the third step, the solution of the second equation in
(5.29) is found for the initial condition
(u )( ) = f (( )) in order to determine
u(x) = (u )(0), where f : S R are the specified data
for u on S.
Single First-Order Quasi-Linear PDE
1.0

0.5

y 0.0

-0.5

-1.0
-3 -2 -1 0 1 2 3
x

Figure 13: Graph and contour plot of a solution u to (5.30) assuming


the boundary values u(x, 0) = 1/(1 + x 2 ) for every x R. In the
contour plot, relatively darker colors refer to relatively smaller
function values.
Single First-Order Quasi-Linear PDE

Example 13
In the following, we find the solution u : R2 R to the
semi-linear equation
u u
+ + u2 = 0 (5.30)
x y

satisfying the boundary condition u(x, 0) = f (x) for every


x R, where f : R R is a differentiable.
Single First-Order Quasi-Linear PDE

Example (cont.)
I For this, let (x, y) R2 .
I The maximal characteristic path satisfying (0) = (x, y)
is given by ( ) = ( + x, + y) for every R.
I This path intersects the data surface R {0} at time
= y.
I In this special case, the right equation in (5.29) is given by

(u ) 0 ( ) = (u )2 ( ) .
Single First-Order Quasi-Linear PDE

Example (cont.)
I The solution of this equation satisfying the initial condition

(u )(y) = f (x y)

is given by

f (x y)
(u )( ) = ,
f (x y)( + y) + 1
Single First-Order Quasi-Linear PDE

Example (cont.)
I where
(
R \ {(y + [f (x y)]1 )} if f (x y) 6= 0
I :=
R if f (x y) = 0 .

I We note that in the case that f (x y) 6= 0 and

y + [f (x y)]1 = 0 , (5.31)

0/ I.
I As a consequence, in this case, there is no C 1 -solution u of
(5.30) on R2 that satisfies u(x, 0) = f (x) for every x R.
Single First-Order Quasi-Linear PDE

Example (cont.)
I In the remaining cases, we conclude that

f (x y)
u(x, y) = (u )(0) = .
1 + yf (x y)
I Since for every not identically vanishing f there is
(x, y) R2 satisfying (5.31),
I we conclude that the only C 1 -solution u of (5.30) on R2 is
given by u(x, y) = 0 for all (x, y) R2 .
Single First-Order Quasi-Linear PDE

Example (cont.)
I On the other hand, there is a differentiable solution u of
(5.30) on defined by

f (x y)
u(x, y) :=
1 + yf (x y)

for every (x, y) ,


I where

:= R2 \ {(x, y) R2 : 1 + yf (x y) = 0} .
Single First-Order Quasi-Linear PDE

In the general, not necessarily semi-linear, case, typically, the


calculation proceeds in two steps.

Steps in the solution of the general quasi-linear case


1) In the first step, a solution (, u ) of (5.29) needs to be
found such that
a) ( ) = x, where x and R is to be determined,
b) (0) is in the data surface,
c) u (0) = f ((0)), where f is a given function on the data
surface prescribing the values of u,
2) Then u(x) = u ( ).
Single First-Order Quasi-Linear PDE

Example 14
In the following, we find a continuously differentiable
u : R, where := (0, ) R, satisfying

u u
u(x, y) (x, y) + x (x, y) = x
x y

for every (x, y) and such that

u(x, 0) = 3x

for x > 0.
Single First-Order Quasi-Linear PDE

Example (cont.)
I As consequence, a and from (5.27) are given by

a(x, y, z) = (z, x) , (x, y, z) = x

for every (x, y) and z R.


I For the solution, let (x, y) .
Single First-Order Quasi-Linear PDE

Example (cont.)
I In this case, the system (5.29) for the characteristic path
: I through (x, y), satisfying ( ) = (x, y) for a to
be determined R,
I and u := u is given by

10 (s) = u (s) , 20 (s) = 1 (s) , u0 (s) = 1 (s)

for every s I,
I where j denotes the composition of the j-th coordinate
projection of R2 and .
Single First-Order Quasi-Linear PDE

Example (cont.)
I The solutions of this system are given by

1 (s) = a cos(s) + b sin(s) , 2 (s) = a sin(s) b cos(s) + c ,


u (s) = a sin(s) + b cos(s)

for every s I := R and a, b, c R.


Single First-Order Quasi-Linear PDE

Example (cont.)
I Following the previous procedure, we arrive at the
following system of equations

a cos( ) + b sin( ) = x , a sin( ) b cos( ) + c = y ,


c = b , b = 3a

for a, b, c, R.
I Hence a and satisfy

a cos( ) 3a sin( ) = x , a sin( ) + 3a cos( ) 3a = y .


Single First-Order Quasi-Linear PDE

Example (cont.)
I As a consequence,

10a sin( ) = y 3x + 3a , 10a cos( ) = x + 3y + 9a ,

I and therefore

a2 6ay x 2 y2 = 0 .

I The latter leads on


p
a = 3y + x 2 + 10y2 .
Single First-Order Quasi-Linear PDE

Example (cont.)
I Hence it follows that

u(x, y) = u ( ) = a sin( ) + b cos( ) = c y = 3a y


p
= 10y 3 x 2 + 10y2 .
Single First-Order Quasi-Linear PDE

I In the following, we describe another approach to the


solution of (5.27).
I For this, we note that the solutions (, u ) of the system
(5.29) are integral curves of the vector field
(a0 , . . . , an+1 ) C 1 ( R, Rn+2 ),
I where an+1 C 1 ( R, R) is defined by

an+1 (x, y) := (x, y)

for every (x, y) R.


I For this reason, there might be suspected a connection
between the solutions of (5.27)
Single First-Order Quasi-Linear PDE
I and the differentiable solutions v : R R of the
linear equation
n+1
X v
ak (x, y) (x, y) = 0 , (5.32)
k=0
xk

(x, y) R.
I Indeed, this is the case. If v is a differentiable solution of
(5.32)
I and u : R is a differentiable function such that

v(x, u(x)) = c

for every x and some c in the range of v,


Single First-Order Quasi-Linear PDE
then
v v u
(x, u(x)) + (x, u(x)) (x) = 0
xk xn+1 xk
for every k {0, . . . , n}, x
and hence
n  
X v v u
0= ak (x, u(x)) (x, u(x)) + (x, u(x)) (x)
xk xn+1 xk
k=0
n
v v X u
= an+1 (x, u(x)) (x, u(x)) + (x, u(x)) ak (x, u(x)) (x)
xn+1 xn+1 xk
k=0
" n #
v X u
= (x, u(x)) ak (x, u(x)) (x) (x, u(x))
xn+1 xk
k=0

for every x .
Single First-Order Quasi-Linear PDE

I If
v
(x, u(x)) 6= 0 ,
xn+1
for every x ,
I the latter implies that u satisfies (5.27).
I We also note that the level surface v1 (c) is a characteristic
surface of (5.32) if (v)(x, y) 6= 0 for every
(x, y) v1 (c).
I Hence we proved the following
Single First-Order Quasi-Linear PDE

Lemma 5.10
Let v : R R be a differentiable solution of the linear
equation
n
X v v
ak (x, y) (x, y) + (x, y) (x, y) = 0 , (5.33)
k=0
xk xn+1

(x, y) R. Further, let c R be in the range of v and


u : R be a differentiable function such that
v
v(x, u(x)) = c , (x, u(x)) 6= 0
xn+1
for all x . Then
Single First-Order Quasi-Linear PDE

Lemma (cont.)
(i) v1 (c) is a characteristic surface of (5.33) if (v)(x, y) 6= 0
for every (x, y) v1 (c) ,
(ii) u satisfies the non-linear equation
n
X u
ak (x, u(x)) (x) = (x, u(x))
k=0
xk

for every x and Graph(u) v1 (c).


Single First-Order Quasi-Linear PDE

Remark 5.10.1
Note that if in addition u is required to satisfy the initial
conditions
u(x) = f (x)
for every x S, where S and f : S R is a given function,
then v needs to satisfy

v(x, f (x)) = c

for every x S.
Single First-Order Quasi-Linear PDE

Further, we note the following.


Lemma 5.11
Let m N, v0 : R R, . . . , vm : R R be
differentiable solutions of (5.33) and f : Rm+1 R be
differentiable. Then v : R R defined by

v(x) := f (v0 (x), . . . , vm (x))

for every x R is a differentiable solution of (5.33).


Single First-Order Quasi-Linear PDE

Proof.
According to the chain rule for vector-valued functions in
several variables, v is differentiable such that
m
v X f vj
(x) = (v0 (x), . . . , vm (x)) (x)
xk j=0
xj xk

for every x R and k {0, . . . , n + 1}.


Single First-Order Quasi-Linear PDE

Proof (cont.).
Hence it follows that
n+1 n+1
" m
#
X v X X f vj
ak (x) (x) = ak (x) (v0 (x), . . . , vm (x)) (x)
k=0
xk k=0 j=0
xj xk
m
" n+1 #
X f X vj
= (v0 (x), . . . , vm (x)) ak (x) (x) = 0 ,
j=0
x j
k=0
x k

where
an+1 (x, y) := (x, y)
for every (x, y) R.
Single First-Order Quasi-Linear PDE
The following gives an example for the application of the
Lemmata 5.10, 5.11.
1.0

0.5

y 0.0

-0.5

-1.0
-1.0 -0.5 0.0 0.5 1.0
x

Figure 14: Graph and contour plot of u from Example 15. In the
contour plot, relatively darker colors refer to relatively smaller
function values.
Single First-Order Quasi-Linear PDE

Example 15
In the following, we find a differentiable u : R, where
:= (0, ) R, satisfying

u u
u(x, y) (x, y) x (x, y) = x
x y

for every (x, y) and such that

u(x, 0) = 2x

for x > 0.
Single First-Order Quasi-Linear PDE

Example (cont.)
For this, we consider differentiable v : R R satisfying
the linear equation
v v v
z (x, y, z) x (x, y, z) + x (x, y, z) = 0 (5.34)
x y z

for all (x, y, z) R.

Two solutions v0 , v1 of this equation are given by

v0 (x, y, z) = x 2 z2 , v1 (x, y, z) = y + z

for every (x, y, z) R.


Single First-Order Quasi-Linear PDE

Example (cont.)
According to Lemma 5.11, for every differentiable f : R2 R,
v : R R defined by

v(x, y, z) := f (v0 (x, y, z), v1 (x, y, z))

for every (x, y, z) R is also a solution of (5.34).


Single First-Order Quasi-Linear PDE

Example (cont.)
In the following, we find one such f satisfying

f ((v0 )(x), (v1 )(x)) = 0

for every x > 0, where

(x) := (x, 0, 2x)

for every x > 0.


Single First-Order Quasi-Linear PDE
Example (cont.)
For this, we notice that

(v0 )(x) = 3x 2 , (v1 )(x) = 2x

and hence that


3
(v0 )(x) + (v1 )2 (x) = 0
4
for every x > 0. As a consequence, f : R2 R, defined by
3 2
f (x, y) := x + y
4
for every (x, y) R2 , is such function.
Single First-Order Quasi-Linear PDE

Example (cont.)
The associated solution v : R R of (5.34) is given by
3
v(x, y, z) = f (v0 (x, y, z), v1 (x, y, z)) = x 2 z2 + (y + z)2
4
1 3 3
= z2 + yz + x 2 + y2
4 2 4
for every (x, y, z) R.
Single First-Order Quasi-Linear PDE

Example (cont.)
By solving the equation

v(x, y, z) = 0 ,

(x, y, z) R, for z, we obtain u : R with the required


properties, p
u(x, y) = 3y + 2 x 2 + 3y2
for every (x, y) , where we note that

v p
(x, y, u(x, y)) = x 2 + 3y2 6= 0
z
for all (x, y) .
Single First-Order Quasi-Linear PDE

As an application, we consider the quasi-linear equation for the


speed distribution u : R2 R of an ideal fluid in one space
dimension under constant pressure and without presence of
external forces24
u u
+u =0. (5.35)
t x

24
The following equation is also referred to as Inviscid Burgers equation.
Single First-Order Quasi-Linear PDE
In this case, a and from (5.27) are given by

a(t, x, y) = (1, y) , (t, x, y) = 0

for every (t, x, y) R3 and (5.29) leads to the following system


of ordinary differential equations for a characteristic path25
:IR

0 ( ) = (1, (u )( )) , (u ) 0 ( ) = 0

for every I which gives

0 ( ) = + 0 (0) , (u )( ) = (u )(0) ,
1 ( ) = (u )(0) + 1 (0) . (5.36)

25
In this case, those paths are particle paths of fluid elements in the fluid
flow.
Single First-Order Quasi-Linear PDE

In particular, u is constant along characteristic curves. We try to


find the value of the solution u at some point (t, x) R2 in
terms of its value at time 0 in some space point that is to be
determined. For this, we demand that

( ) = (t, x) , 0 (0) = 0 .
Single First-Order Quasi-Linear PDE

Then it follows from (5.36) that = t and

u(t, x) = (u )( ) = (u )(0) = u(0, 1 (0))


= u(0, 1 ( ) (u )( ) ) = u(0, x u(t, x) t)

which gives the implicit representation of u(t, x) in terms of the


data u0 := u(0, ) at time zero

u(t, x) = u0 (x u(t, x) t) (5.37)

for all t, x R.
Single First-Order Quasi-Linear PDE

Indeed, if u0 : R R is differentiable and u : R is a


partially differentiable solution of (5.37) on a non-empty open
subset of R2 , then it follows for every (t, x) that
 
u u
(t, x) = u(t, x) + t (t, x) u00 (x u(t, x) t) ,
t t
 
u u
(t, x) = 1 t (t, x) u00 (x u(t, x) t) .
x x
Single First-Order Quasi-Linear PDE
This implies that
u
[1 + t u00 (x u(t, x) t)] (t, x) = u(t, x) u00 (x u(t, x) t) ,
t
u
[1 + t u00 (x u(t, x) t)] (t, x) = u00 (x u(t, x) t) (5.38)
x
and hence that
 
u u
[1 + t u00 (x u(t, x) t)] +u (t, x) = 0
t x

which leads to  
u u
+u (t, x) = 0 .
t x
Single First-Order Quasi-Linear PDE

Here, it has been used that

1 + t u00 (x u(t, x) t) 6= 0 .

Otherwise, it would follow by the second equation in (5.38) that

u00 (x u(t, x) t) = 0

which would lead to the contradiction that 1 = 0.


Single First-Order Quasi-Linear PDE

We consider the special case that

u(0, x) = x

for every x R. Then (5.37) leads to

u (t, x) = [ x u (t, x) t ]

and hence
x
u (t, x) =
t1
for all
(t, x) := ( R \ {1} ) R .
Single First-Order Quasi-Linear PDE

Indeed, u C 1 ( , R) and
 
u u x x 1
+ u (t, x) = + =0
t x (t 1)2 t 1 t 1

for all (t, x) .


Single First-Order Quasi-Linear PDE

As a consequence, u has regular initial data at t = 0 but is


singular at t = 1.

To shed light on the origin of this behavior, we calculate the


characteristic paths corresponding to u .

The characteristic vector field a is given by


 
x
a(t, x) = 1,
t1

for every (t, x) := (R \ {1}) R.


Single First-Order Quasi-Linear PDE

Hence the maximal characteristic path = (0 , 1 ) : I


such that (0) = (0, x) satisfies
 
0 1 ( )
( ) = 1,
0 ( ) 1

for every I. Therefore, I = (, 1) and

( ) = (, x (1 ))

for every I.
Single First-Order Quasi-Linear PDE

As a consequence the corresponding characteristic curves all


intersect in the point (1, 0)!
Single First-Order Quasi-Linear PDE
t

t =1

x
- 10 -6 -2 0 2 6 10

Figure 15: Characteristic curves corresponding to u . u is singular


for all x R at t = 1.
Single First-Order Quasi-Linear PDE

Note that : R R2 , defined by

( ) = (, x (1 ))

for every R, is characteristic also for > 1.


Single First-Order Quasi-Linear PDE

In the following, it will be shown that a singular behavior of


solutions corresponding to strictly decreasing data such as u
is to be expected. Note that for physical reasons, both solutions
u have to be discarded because of the unlimited growth of the
data for x .
Single First-Order Quasi-Linear PDE

Singular behavior of the solutions has to expected if


characteristic curves intersect in a point. In general, u cannot be
univalent in such a point. In such a case the solution is said to
develop a shock. In the following, we explore whether such
intersection is possible.
Single First-Order Quasi-Linear PDE

For this, we define : R2 R2 by

(t, x) := (t, x + u(0, x) t)

for every (t, x) R2 . Then for every x R the corresponding


(, x) is such that (0, x) = (0, x).
Single First-Order Quasi-Linear PDE

Indeed, two such characteristic curves starting from points


(0, x), (0, x 0 ) R2 , where x 0 < x, intersect at time

x x0
t :=
u(0, x) u(0, x 0 )
if
u(0, x) < u(0, x 0 ) ,
that is, if the speed u(0, x 0 ) of the fluid element starting from
x 0 < x at time 0 is larger than the speed of the fluid element
starting from x at time 0. Then the former fluid element will
overtake the latter fluid element.
Single First-Order Quasi-Linear PDE

As a consequence, such intersection is impossible for strictly


increasing u(0, ). On the other hand, solutions corresponding
to non-vanishing C data with compact support will develop
shocks.

As a consequence, the demand that u is C 1 is too strong!


Single First-Order Quasi-Linear PDE
t

GHL

t2

R1


R2

t1 R

x
Ht 1L Ht 2L

Figure 16: Paths of integration in the derivation of the shock condition


(5.40).
Single First-Order Quasi-Linear PDE

In the following, we define a class of weak solutions for (5.35).


For this, let R, S C 1 (R, R) such that

S 0 (x) = x R 0 (x)

for every x R.26 Further, let u C 1 (R2 , R) be a solution of


(5.35).

26
For instance, R(x) := x and S(x) := x 2 /2 for all x R.
Single First-Order Quasi-Linear PDE

Then
(R u) (S u) u u
+ = (R 0 u) + (S 0 u)
t x t
  x
u u
= (R 0 u) +u =0
t x

which brings (5.35) into the form of a conservation law.


Single First-Order Quasi-Linear PDE

Further, let be some element of C 1 (R2 , R) that vanishes


outside some compact interval. Then it follows by partial
integration (which is a consequence of Greens theorem)
 
(R u) (S u)
Z
0= + dtdx
R2 t x
Z  

= (R u) + (S u) dtdx .
R2 t x
Single First-Order Quasi-Linear PDE

In the following, we say that an almost everywhere continuous


u : R2 R is a weak solution of (5.35) if
Z  

(R u) + (S u) dtdx = 0
R2 t x

for all elements of C 1 (R2 , R) that vanishes outside some


compact interval.
Single First-Order Quasi-Linear PDE

For clarification, we consider in the following the simplified


situation that u C 1 (R2 \ G(), R) satisfies (5.35) on its
domain where C 1 (R, R) is strictly increasing and surjective.
Here G() := {(t, (t)) : t R}.

In addition, let u be such that

lim u(t, x) = u (t, (t)) , lim u(t, x) = u+ (t, (t))


x(t) , x<(t) x(t) , x>(t)
(5.39)
for every t R and some u , u+ C(G(), R).
Single First-Order Quasi-Linear PDE
In the following, we investigate whether u is a weak solution of
(5.35). For this, let an element of C 1 (R2 , R) that vanishes
outside a compact interval

R := [t1 , t2 ] [(t1 ), (t2 )] .

In addition, we define
[ [
R1 := ( {t} [(t1 ), (t)] ) , R2 := ( {t} [(t), (t2 )] ) .
t[t1 ,t2 ] t[t1 ,t2 ]

Then it follows by Greens theorem that 27

27
Note that in Fig 16 the convention is used that the first coordinate is
marked on the vertical axis, whereas the second coordinate is marked on the
horizontal axis.
Single First-Order Quasi-Linear PDE

Z  

(R u) + (S u) dtdx
R t x
2 Z  
X
= (R u) + (S u) dtdx
i=1 Ri
t x
2 Z   
X [(R u)] [(S u)] (R u) (S u)
= + +
i=1 Ri
t x t x
2  
[(R u)] [(S u)]
X Z
= + dtdx
i=1 Ri
t x
X2 Z
= ((S u), (R u)) d(t, x)
i=1 Ri
Single First-Order Quasi-Linear PDE

Z t2
= (t, (t)) [S(u (t, (t))) R(u (t, (t))) 0 (t)] dt
t
Z t12
+ (t, (t)) [S(u+ (t, (t))) R(u+ (t, (t))) 0 (t)] dt
t1

Since is arbitrary otherwise, it follows that u is a weak


solution of (5.35) if and only if the following jump (or shock)
condition is satisfied

S(u+ (t, (t))) S(u (t, (t)))


= [R(u+ (t, (t))) R(u (t, (t)))] 0 (t) (5.40)

for almost every t R.


Single First-Order Quasi-Linear PDE

This condition connects the speed of the propagation of the


discontinuity with the amounts by which R and S jump across
the discontinuity. Note that this condition is satisfied if u is
continuous. In the following, it is also used for the definition of
weak solutions of (5.35) on a non-trivial open subset of R2
satisfying (5.39) for a continuously differentiable path in .
Single First-Order Quasi-Linear PDE
t

x
-4 -2 0

Figure 17: Curve along which u from Problem 16 develops a shock.


Single First-Order Quasi-Linear PDE

Example 16
Decide whether
( 
23 t + 3x + t 2 for 4x + t 2 > 0
u(t, x) :=
0 for 4x + t 2 < 0

for all (t, x) := ((0, ) R) \ {(t, (t)) : t > 0} is a weak


solution of (5.35) for R(x) := x and S(x) := x 2 /2 for all x R.
Here (t) := t 2 /4 for all t (0, ).
Single First-Order Quasi-Linear PDE

Example (cont.)
Solution:
I First, we notice that it follows for every (t, x) R2 that
4x + t 2 > 0 implies that x > t 2 /4
I and hence that 3x + t 2 > t 2 /4.

I As a consequence, u is well-defined.
Single First-Order Quasi-Linear PDE

Example (cont.)
I Obviously, u C 1 (, R) and
   
u u 2 t
+u (t, x) = 1+
t x 3 3x + t 2

 
2 2 3
(t + 3x + t 2 )
3 3 2 3x + t 2
 
2 t t
= 1+ 1 =0
3 3x + t 2 3x + t 2

for (t, x) such that 4x + t 2 > 0.


Single First-Order Quasi-Linear PDE

Example (cont.)
I In addition,
t 1
(t)2
= 0 (t) = 2
2 t
for t > 0.
I Hence u also satisfies the shock condition (5.40)
I and is a weak solution of (5.35) on .
I Note that also u has the unphysical feature that u(t, ) is
unbounded for all t > 0.
Single First-Order Quasi-Linear PDE

Figure 18: Graph of the solution u of Eulers equation from


Problem 17 for a = 1, b = 0.3. The graph extends beyond the bound
(5.44) ( 0.6415) for t.
Single First-Order Quasi-Linear PDE

The following example finds a solution to (5.35) corresponding


to initial data that decay at infinity.
Example 17
Find a solution u of (5.35) to the initial data

2b
u(0, x) =
x2 + a2
for all x R, where a > 0, b R .
Single First-Order Quasi-Linear PDE

Example (cont.)
Solution:
I Such a solution is given by a solution of the implicit
equation
2b
u= . (5.41)
(x ut)2 + a2
I Here for simplicity of notation, the symbol u is used
instead of u(t, x).
I The introduction of v := ut x leads for t 6= 0 on
v+x 2b
= 2
t v + a2
Single First-Order Quasi-Linear PDE

Example (cont.)
I which is equivalent to the cubic equation

0 = (v+x)(v2 +a2 )2bt = v3 +xv2 +a2 v+a2 x2bt (5.42)

I We bring 5.42 into a form which will allow us the


application of Cardanos formulas.
I This is achieved by introduction of
x
w := v +
3
Single First-Order Quasi-Linear PDE
Example (cont.)
I which leads on

0 = v3 + xv2 + a2 v + a2 x 2bt
 x 3  x 2  x
= w +x w + a2 w + a2 x 2bt
3 3 3
3 2 x2 x3 2 2x 2 x3
= w xw + w + xw w + + a2 w
3 27 3 9
2a2
+ x 2bt
3
1 2 3
= w3 + (3a2 x 2 )w + (x + 9a2 x 27bt)
3 27
= w3 + 3pw + 2q , (5.43)
Single First-Order Quasi-Linear PDE

Example (cont.)
I where
1 1 3
p := (3a2 x 2 ) , q := (x + 9a2 x 27bt) .
9 27
I According to Cardanos formulas, a solution of the last
cubic equation is given by
 p 1/3  p 1/3
w = q + q2 + p3 + q q2 + p3 .
Single First-Order Quasi-Linear PDE

Example (cont.)
I w is real if
( a/ 3 )3
|t| 6 (5.44)
|b|
since q2 + p3 is positive for these cases.
I The last can be seen as follows. First,

36 (q2 + p3 ) = (x 3 + 9a2 x 27bt)2 + (3a2 x 2 )3 > 0

for
|x| 6 a 3 .
Single First-Order Quasi-Linear PDE

Example (cont.)
I Further,

36 (q2 + p3 ) = (x 3 + 9a2 x 27bt)2 + (3a2 x 2 )3


= x 6 + 81a4 x 2 + 729b2 t 2 + 18a2 x 4 54btx 3 486a2 btx
+ 27a6 27a4 x 2 + 9a2 x 4 x 6
= 27a2 x 4 54btx 3 + 54a4 x 2 + 729b2 t 2 486a2 btx + 27a6 .

I Hence, q2 + p3 > 0 for t = 0.


Single First-Order Quasi-Linear PDE

Example (cont.)
I In the following,
we consider the cases t 6= 0 and
|x| > a 3.
I Then

36 (q2 + p3 ) = 27(a2 s2 2bs) s2 t 4 + 54a2 (a2 s2 9bs) t 2


+ 729b2 t 2 + 27a6 ,

where s := x/t.
Single First-Order Quasi-Linear PDE

Example (cont.)
I Hence q2 + p3 > 0 if |s| > 9|b|/a2 or equivalently if

9|b|
|x| > |t| .
a2
I Hence q2 + p3 > 0 if (5.44) is satisfied.
I Hence it follows that u : \ ({0} R) R,
I where " #
( a/ 3 )3 ( a/ 3 )3
:= , R ,
|b| |b|
Single First-Order Quasi-Linear PDE

Example (cont.)
I defined by

1
u(t, x) := 2x (5.45)
3t
h p i1/3
+ (x 3 + 9a2 x 27bt) + (x 3 + 9a2 x 27bt)2 + (3a2 x 2 )3
h p i1/3 
3 2 3 2 2 2 2 3
+ (x + 9a x 27bt) (x + 9a x 27bt) + (3a x )

for every (t, x) \ ({0} R),


I is a solution of the implicit equation (5.41) on
\ ({0} R).
Single First-Order Quasi-Linear PDE

Example (cont.)
I Further, it is not difficult to see that

lim tu(t, x) = 0
t0

for every x R.
I Hence it follows by (5.41) that u has a continuous extension
to {0} R given by (R R, x 7 2b/(x 2 + a2 )).
I In particular, it follows that this continuous extension is a
weak solution of (5.35) on .
Single First-Order Quasi-Linear PDE

1) Using the characteristic method, find a differentiable u


satisfying the given conditions.
u u
a) (x, y) + (x, y) + u2 (x, y) + 1 = 0 , (x, y) R2 ,
x y
u(x, 3x) = 1/(1 + x 2 ) for x R ,
u u
b) xu(x, y) (x, y) + yu(x, y) (x, y) = xy , (x, y) (0, )2 ,
x y
2
u(x, 1 x) = x for x (0, ) ,
u u p
c) x (x, y) + y (x, y) = 2xy 1 u2 (x, y) ,
x y
(x, y) (0, ) R , u(x, y) = 1/2 for all (x, y) S 1 (0, )2 ,
Single First-Order Quasi-Linear PDE

1)
u u
d) (t, x) + u(t, x) (x, y) = 1 , (t, x) (2, 2) (1, ) ,
t x
u(0, x) = x for x (1, ) ,
u u
e) (t, x) + u(t, x) (x, y) = x , (t, x) R2 , u(0, x) = 1 for x R ,
t y
u u
f) (t, x) + u(t, x) (x, y) = x , (t, x) (/2, /2) R ,
t y
u(0, x) = 1 for x R ,
u u
g) (t, x) + u(t, x) (x, y) = u3 (x, y) , (t, x) (0, ) R ,
t y
u(0, x) = 1/x for x > 0 .
Single First-Order Quasi-Linear PDE

2) Use Lemmata 5.10, 5.11 to find the solution of the initial


value problem from
a) Example 14 ,
b) Problem 1b) ,
c) Problem 1d) .
3) Use Lemmata 5.10, 5.11 to derive the implicit form (5.41)
of Eulers equation (5.35).
First-Order Quasi-Linear Systems of PDE

Such a system is of the form


n
X u
Ak (x, u(x)) (x) = (x, u(x)) (6.1)
k=0
xk

for every x , where n, p N , is a non-empty open subset


of Rn+1 , A0 , . . . , An C 1 ( Rp , M(p p, R)),
B C( Rp , M(p p, R)), C( Rp , Rp ) are given,
and u : Rp is the unknown map which is assumed to be in
C 1 (, Rp ). In addition, denotes matrix multiplication.
First-Order Quasi-Linear Systems of PDE

Particularly important for applications is the case that A1 , . . . , An


are all assuming values in the subsapce of symmetric matrices.
In this case, the system is called symmetric hyperbolic.
First-Order Quasi-Linear Systems of PDE

I In a first step, we investigate the solubility of (6.1) for the


derivative ( )u into the normal direction
: S Rn+1 \ {0} of a regular C 1 -hypersurface S that is
contained in ,
I where n
X
( )u := [( )ui ] ei
i=0

and n
X g
( )g := k
k=0
xk
for every differentiable g : R.
First-Order Quasi-Linear Systems of PDE
I It follows that
n n
X u X
Ak (x, u(x)) (x) = Ak (x, u(x)) [(ek )u](x)
k=0
xk k=0
n    
X k k
= Ak (x, u(x)) ek 2 + 2 u (x)
k=0
|| ||
n
!
1 X
= k (x).Ak (x, u(x)) [( )u](x)
|(x)|2 k=0
n
X
+ Ak (x, u(x)) [(ek ) u] (x)
k=0
for every x S,
I where
k
ek := ek
||2
for every k {0, . . . , n}.
First-Order Quasi-Linear Systems of PDE
I Note that ek (x) is tangential to S in x and hence that
[(ek ) u] (x)
is a derivative tangential to S
I which can be calculated alone from data for u given on S.
I Hence (6.1) is everywhere on S solvable for ( )u if and
only if !
Xn
det k (x).Ak (x, u(x)) 6= 0
k=0
for all x S.
I In this case, we say that S is non-characteristic for u.
I and, by help of (6.1),
I the derivative of u in every point of S into the normal
direction of S can be calculated from the knowledge of the
values of u on S.
First-Order Quasi-Linear Systems of PDE

I On the other hand, if


n
!
X
det k (x).Ak (x, u(x)) =0
k=0

for all x S,
I we say that S is characteristic for u.
First-Order Quasi-Linear Systems of PDE

As a consequence, non-characteristic regular C 1 -hypersurfaces


are possible candidates for the specification of free data for the
solutions of (5.4), whereas data cannot be freely specified for
those solutions on characteristic surfaces.

Since the differentiability of u does not enter the previous


definitions, we extend these to all solutions of (6.1).
First-Order Quasi-Linear Systems of PDE
Definition 6.1
For a, possibly weak, solution u of (5.4), we define
(i) S is non-characteristic for u if
n
!
X
det k (x).Ak (x, u(x)) 6= 0
k=0

for all x S.
(ii) S is characteristic for u if
n
!
X
det k (x).Ak (x, u(x)) =0
k=0

for all x S.
First-Order Quasi-Linear Systems of PDE

The following example shows that, differently to the case of a


quasi-linear first-order PDE for one real-valued function,
characteristic hypersurfaces need not exist for solutions of
quasi-linear systems.
First-Order Quasi-Linear Systems of PDE
Example 18
(Cauchy-Riemann equations) The real part u and imaginary
part v of a holomorphic function defined on a non-trivial open
subset of R2 satisfy
u v u v
= , = . (6.2)
x y y x
An equivalent form of (6.2) is given by
       
1 0 0 1 u 0
+ = , (6.3)
0 1 x 1 0 y v 0

Hence it follows for every vector (n1 , n2 ) R2 \ {(0, 0)} that


      
1 0 0 1 n1 n2
det n1 . + n2 . = det = |n|2 > 0 .
0 1 1 0 n2 n1
First-Order Quasi-Linear Systems of PDE
Example (cont.)
I Hence there are no characteristics for the system (6.3)
I and every regular hypersurface contained in is
non-characteristic.
I As a consequence, it might be suspected that any regular
hypersurface is a suitable data surface for (18).
I This is not the case. In general, the initial value problem
for (18) on regular hypersurfaces is not well-posed.
I In the following, we indicate this fact, only.
I For this, we use that, as is shown in complex analysis in one
variable, any differentiable map (u, v) : R2 defined on
some non-trivial open subset R2 whose component
functions satisfy (6.3) is infinitely often differentiable.
First-Order Quasi-Linear Systems of PDE

Example (cont.)
I As a consequence, such (u, v) satisfies

2u 2v 2v 2u
= = = 2 ,
x 2 xy yx y
2 2 2
v u u 2v
= = =
x 2 xy yx y2
I and hence
2u 2u 2v 2v
+ = + =0.
x 2 y2 x 2 y2
First-Order Quasi-Linear Systems of PDE

Example (cont.)
I Generally, the initial value problem for the last equations is
not well-posed.
I Later, this will be indicated in the Example 31 by
Hadamard.
First-Order Quasi-Linear Systems of PDE

In the following, we consider the special case n = 1. In this


case, characteristic surfaces (so far existent) are 1-dimensional
and hence curves. Hence it might be expected that restriction of
(6.1) to characteristic paths28 leads on a ordinary differential
equation. Indeed, this is true as we will see in the following.

28
i.e., parametrizations of such curves
First-Order Quasi-Linear Systems of PDE
I For this, we assume that there is a characteristic regular
C 1 -hypersurface S for (6.1) with normal field
: S Rn \ {0}.
I Further, let : I S be differentiable, where I is a
non-trivial open interval of R.
I Then

ek (( )) 0 ( ) = ek (( )) 0 ( ) = k0 ( )

for every I and k {0, . . . , n}.


I In addition, let w : S Rp such that
n
!
X
w(x) ker k (x).Ak (x, u(x)) (6.4)
k=0

for all x S.
First-Order Quasi-Linear Systems of PDE

I Note that
n
!
X
det k (x).Ak (x, u(x))
k=0
n
!
X
= det k (x).Ak (x, u(x)) =0
k=0

for every x S.
First-Order Quasi-Linear Systems of PDE
I Hence such w exists. Then
n
X u
w(x) Ak (x, u(x)) (x)
xk
k=0
n
!
1 X
= w(x) k (x).Ak (x, u(x)) [( )u](x)
|(x)|2
k=0
n
X
+ w(x) Ak (x, u(x)) [(ek ) u] (x)
k=0
" n
! #
1 X
= k (x).Ak (x, u(x)) w(x) [( )u](x)
|(x)|2
k=0
n
X
+ w(x) Ak (x, u(x)) (ek ) u
k=0
n
X
= w(x) Ak (x, u(x)) [(ek ) u] (x)
k=0

for every x S.
First-Order Quasi-Linear Systems of PDE

I Using for the first time that n = 1, it follows that

0 ( ) ek (( )) 0 k0 ( ) 0
ek (( )) = ( ) = ( ) ,
| 0 ( )|2 | 0 ( )|2
1
X u
[(ek ) u] (( )) = [ek (( ))]i (( ))
i=0
xi
1
k0 ( ) X 0 u k0 ( )
= 0 ( ) (( )) = (u )0 ( )
| ( )|2 i=0 i xi | 0 ( )|2

for k = 0, 1.
First-Order Quasi-Linear Systems of PDE

I Hence (6.1) implies the ordinary differential equation


1
X k0 ( )
0 ( )|2
.(w )( ) Ak (( ), (u )( )) (u )0 ( )
k=0
|
= (w )( ) (( ), (u )( )) (6.5)

for every t I.
First-Order Quasi-Linear Systems of PDE

The PDE in Theorem 5.9 is a particular case of a


inhomogeneous linear symmetric hyperbolic system. The
following is a natural generalization of Theorem 5.9 to more
general inhomogeneous linear symmetric hyperbolic systems of
PDE. The study of the latter systems is continued in Section 3.
As a by product of this Section, we arrive at the motivation
behind the mutiplication of the equation (5.25) by the
multiplicator (R2 R, (t, x) 7 2(t x)2n (u)(t, x)) in the
derivation of the conservation law (5.26) of Example 12.
First-Order Quasi-Linear Systems of PDE

Theorem 6.2
Let n {1, 2, 3}, p N \ {0, 1}, T > 0, S a nonempty bounded
open subset of Rn to which Gauss theorem is applicable and
be an open subset of Rn+1 containing [0, T ] S. In addition, let
A1 , . . . , An C 1 (, M(p p, R)) with values in the subspace of
symmetric matrices, B C(, M(p p, R)), C(, Rp )
such that for every (t, x) (0, T ) S the corresponding
symmetric matrix
n
X Ak
B(t, x) + (B(t, x)) (t, x) (6.6)
k=1
xk

is in particular positive.
First-Order Quasi-Linear Systems of PDE

Theorem (cont.)
Finally, let u, v C 1 (, Rp ) such that
n n
u X u v X v
+ Ak +Bu = + Ak +Bv = ,
x0 k=1 xk x0 k=1 xk

P(t,x) u(t, x) = P(t,x) v(t, x) = 0 , (6.7)


for every (t, x) of (0, T ) S,
First-Order Quasi-Linear Systems of PDE

Theorem (cont.)
where P(t,x) is the orthogonal projection onto the span of
eigenvectors of
Xn
k (x)Ak (t, x)
k=1

corresponding to strictly negative eigenvalues, where


: S Rn is the outer unit normal field of S, and
u|{0}S = v|{0}S . Then u|[0,T ]S = v|[0,T ]S .
First-Order Quasi-Linear Systems of PDE
Proof.
In the following, we use the notation h | i for the Euclidean
scalar product on Rp . Further, we define w := u v. Then
n
w X w
+ Ak +Bw = 0
x0 k=1 xk

and hence
n
w X w
0 = 2 hw| i+ 2 hw|Ak i + 2 hw|B wi
x0 xk
k=1
n n
hw|wi X hw|Ak wi AkX
= + wi + hw|B wi + hw|B wi
hw|
x0 xk xk
k=1 k=1
n n
!
hw|wi X hw|Ak wi
X Ak
= + + hw| B + B wi .
x0 xk xk
k=1 k=1
First-Order Quasi-Linear Systems of PDE

Proof (cont.).
As a consequence,
(div q)(t, x) 6 0
for all (t, x) (0, T ) S, where

q := (hw|wi , hw|A1 wi , . . . , hw|An wi) .


First-Order Quasi-Linear Systems of PDE
Proof (cont.).
Further,
n
X
(q1 (t, x), , qn (t, x)) (x) = k (x) hw(t, x)|Ak (t, x)w(t, x)i
k=1
n
X
= hw(t, x)| k (x)Ak (t, x)w(t, x)i
k=1
n
X
= h(E P(t,x) )w(t, x)| k (x)Ak (t, x)(E P(t,x) )w(t, x)i
k=1
n
X
= hw(t, x)|(E P(t,x) ) k (x)Ak (t, x)(E P(t,x) )w(t, x)i > 0
k=1

for all t (0, T ) and x S.


First-Order Quasi-Linear Systems of PDE

Proof (cont.).
Hence according to Lemma 5.7
Z
hw|wi dvn 6 0
S

for all t [0, T ] which implies that w(t, x) = 0 for all


(t, x) [0, T ] S.
First-Order Quasi-Linear Systems of PDE
Remark 6.2.1
The boundary condition (6.7) is called maximal dissipative.
Remark 6.2.2
Note that if u : Rp satisfies
n
u X u
+ Ak +Bu = ,
x0 k=1 xk

then v : Rp , defined by v(t, x) := et u(t, x) for every


(t, x) , satisfies
n
v X v
+ Ak + (B + E) v = ,
x0 k=1 xk

where E M(p p, R) denotes the unit matrix.


First-Order Quasi-Linear Systems of PDE

Remark (cont.)
Hence the statement of Theorem 5.9 is still true if (5.24) is
replaced by the condition that there is R such that the
symmetric matrix
n

X Ak
B(t, x) + (B(t, x)) (t, x) + E
k=1
xk

is in particular positive for every (t, x) ((0, T ) S.


First-Order Quasi-Linear Systems of PDE

I It remains the question of how to find suitable candidates


for vector fields q for the application Lemma 5.7 in the
case of symmetric hyperbolic systems of PDE.
I For this, let n {1, 2, 3}, p N , be an open subset of
Rn+1 , A1 , . . . , An C 1 (, M(p p, R)) with values in the
subspace of symmetric matrices,
I B C(, M(p p, R)) and u, v C 1 (, Rp ) such that
n n
u X u v X v
+ Ak +Bu = + Ak +Bv = 0 .
x0 k=1 xk x0 k=1 xk
First-Order Quasi-Linear Systems of PDE

I Then
n
u X u
0 =h |vi + hAk |vi + hB u|vi
x0 k=1
x k
n
v X v
+ hu| i+ hu|Ak i + hu|B vi
x0 k=1
xk
n n
hu|vi X hu|Ak vi X Ak
= + + hu|(B + B ) vi ,
x0 k=1
xk k=1
xk

I where we use the notation h | i for the Euclidean scalar


product on Rp .
First-Order Quasi-Linear Systems of PDE
I Hence
n

X Ak
div q(u, v) = hu|(B + B ) vi ,
k=1
xk

I where

q(u, v) := (hu|vi , hu|A1 vi , . . . , hu|An vi) .

I In particular, if S : C 1 (, Rp ) C 1 (, Rp ) is a symmetry
transformation associated with the system,
I i.e., a map that transforms solutions of the system into
solutions,
I then q(u, Su) is a candidate for the application of
Lemma 5.7.
First-Order Quasi-Linear Systems of PDE

I For instance in the case that p = 1, A1 , . . . , An are constant


and B = 0,
n
vu X vu
+ Ak
x0 k=1
xk
n
! n
!
v X v u X u
= + Ak u+v + Ak
x0 k=1 xk x0 k=1 xk

I and hence Sv : C 1 (, Rp ) C 1 (, Rp ) defined by


Su := vu is a symmetry transformation.
I This has been exploited in Example 12.
First-Order Quasi-Linear Systems of PDE

I We consider compressible Eulers equations in one space


dimension
I under assumption of a polytropic equation of state and
absence of external forces.
First-Order Quasi-Linear Systems of PDE
I The system for the velocity distribution u : R2 R and
the density : R2 (0, ) of an ideal fluid in one space
dimension without presence of external forces and under
the assumption of a polytropic equation of state is given by
u u 1 p
+u = ,
t x x
u
+ =0,
t x
 

p = p0 , (6.8)
0
I where
1
=1+ (> 1) ,
n
n N is the so called polytropic index of the fluid and
and p0 , 0 > 0.
First-Order Quasi-Linear Systems of PDE
I The system is quasi-linear.
I Introduction of the new dimensionless variable := /0
gives the system
u u c2
+u + =0,
t x x
u
+ +u =0, (6.9)
t x x
I where c is the velocity of sound in the fluid
 1/2
(1)/2 p0
c := c0 , c0 := . (6.10)
0
I or equivalently
       
1 0 u c2 / u 0
+ = . (6.11)
0 1 t u x 0
First-Order Quasi-Linear Systems of PDE
I A normal field to a regular characteristic curve satisfies
the equation
 
0 + 1 u 1 c2 /
0 = det = (0 + 1 u)2 12 c2
1 0 + 1 u
= [0 + 1 (u + c)] [0 + 1 (u c)]

I and hence
0 + 1 (u c) = 0 .
I Therefore, the integral curves of

(1, u c)

are characteristic paths.29


29
Note that, equaling the number of unknown fields, there are two
characteristic vector fields in this case.
First-Order Quasi-Linear Systems of PDE
I Multiplication of the second equation of (6.9) by c/ leads
to
u u c
+u + c =0,
t x x
 
u c
c + +u =0.
x t x
I Addition of the last equations and subtraction of the second
from the first equation leads to a system that includes only
derivatives in the directions of the characteristic fields
   
c
+ (u + c) u+ + (u + c) = 0 ,
t x t x
   
c
+ (u c) u + (u c) = 0 .
t x t x
(6.12)
First-Order Quasi-Linear Systems of PDE

+ -

x x x

Figure 19: Characteristic curves through the point (t, x) corresponding


to the characteristic paths , + . Here x 0 := (0) and
x 00 := + (0).
First-Order Quasi-Linear Systems of PDE
I If + , are integral curves of (1, u + c) and (1, u c),
I the latter equations lead to ordinary differential equations
along + , of the type (6.5):
 
+ 0 c
(u ) + +
( + )0

 
0 c
= (u ) ( )0 = 0 .

I The latter equations give rise to so called simple wave
solutions that propagate only along one characteristic
direction by the assumption that

u = F (6.13)

for some function F which is to be determined.


First-Order Quasi-Linear Systems of PDE
I Substitution of this ansatz into the first equation of (6.12)
gives
 
+ 0 c
(u ) + +
( + )0

 
+ 0 c
= (F ) + +
( + )0

  
0 c
= F + ( + )0 = 0
+

if
2c0 (1)/2
F(x) := x
1
for all x > 0.
First-Order Quasi-Linear Systems of PDE
I Note that this implies that
2
u= c.
1
I The second equation in (6.12) is satisfied if

= (0) . (6.14)
I As a consequence, the characteristic path satisfies the
system of ordinary differential equations
 
0 +1
( ) = (1, (u c) )( ) = 1, (c ) ( )
1
 
+1
= 1, (c )(0)
1
for every of its domain I.
First-Order Quasi-Linear Systems of PDE

I This gives
+1
0 ( ) = +0 (0) , 1 ( ) = (c )(0) +1 (0) .
1
(6.15)
I We try to find the value of the solution at some point
(t, x) R2 in terms of its value at time 0 in some space
point to be determined.
I For this, we demand that

( ) = (t, x) , 0 (0) = 0 .
First-Order Quasi-Linear Systems of PDE

I Then it follows from (6.15) that = t


I and from (6.14) that

(t, x) = ( )( ) = ( )(0) = (0, 1 (0))


 
+1
= 0, 1 ( ) + (c )(0)
1
 
+1
= 0, x + c(t, x) t . (6.16)
1
First-Order Quasi-Linear Systems of PDE
I Substitution of (6.13) into the second equation of (6.12)
gives
 
+ 0 c
(u ) +
( + )0

 
+ 0 c
= (F ) +
( + )0

  
0 c
= F ( + )0 = 0
+

I if
2c0 (1)/2
F(x) := x
1
for all x > 0.
First-Order Quasi-Linear Systems of PDE
I Note that this implies that
2
u= c.
1
I The second equation in (6.12) is satisfied if

+ = + (0) . (6.17)
I As a consequence, the characteristic path + satisfies the
system of ordinary differential equations
 
+0 + +1 +
( ) = (1, (u + c) )( ) = 1, (c ) ( )
1
 
+1 +
= 1, (c )(0)
1
for every of its domain I.
First-Order Quasi-Linear Systems of PDE

I This gives
+1
0+ ( ) = +0+ (0) , 1+ ( ) = (c + )(0) +1+ (0) .
1
(6.18)
I We try to find the value of the solution at some point
(t, x) R2 in terms of its value at time 0 in some space
point to be determined.
I For this, we demand that

+ ( ) = (t, x) , 0+ (0) = 0 .
First-Order Quasi-Linear Systems of PDE

I Then it follows from (6.18) that = t and from (6.17) that

(t, x) = ( + )( ) = ( + )(0) = (0, 1+ (0))


 
+ +1 +
= 0, 1 ( ) (c )(0)
1
 
+1
= 0, x c(t, x) t . (6.19)
1
First-Order Quasi-Linear Systems of PDE

More generally, we can find an implicit representations of more


general solutions of (6.12) by using so called Riemann
invariants of this system. Since
c
= c0 (3)/2 ,

it follows that
   
c 2c0 (1)/2
+ (u c) = + (u c)
t x t x 1
 
2
= + (u c) c.
t x 1
First-Order Quasi-Linear Systems of PDE

Hence the system (6.12) is equivalent to the system


  
2
+ (u + c) u+ c =0,
t x 1
  
2
+ (u c) u c =0.
t x 1

The functions u (2c/( 1)) are Riemann invariants for the


system (6.12), i.e., functions of the unknowns that are constant
along one family of characteristic paths.
First-Order Quasi-Linear Systems of PDE
This implies that
    
2 2 +1
u+ c (t, x) = u + c 0, x c(t, x) t
1 1 1
    
2 2 +1
u c (t, x) = u c 0, x + c(t, x) t
1 1 1
and hence the implicit representations
  
1 2 +1
u(t, x) = u+ c 0, x c(t, x) t
2 1 1
  
2 +1
+ u c 0, x + c(t, x) t (6.20)
1 1
  
1 2 +1
c(t, x) = u+ c 0, x c(t, x) t
4 1 1
  
2 +1
u c 0, x + c(t, x) t .
1 1
First-Order Quasi-Linear Systems of PDE

Problem 6.3
Calculate the solutions of (6.20) under the assumption of linear
u(0, ) and (0, ). In particular, verify whether the results
satisfy (6.8).
First-Order Quasi-Linear Systems of PDE

I For motivation of the definition of hyperbolic systems of


quasi-linear first-order PDE,
I we consider the special case of (6.1) that = Rn ,
I A0 equals the unit matrix E, A1 , . . . An are constant matrices
and of vanishing B and :
n
u X u
(x) + Ak (x) = 0 (6.21)
x0 k=1
xk

for every x Rn .
First-Order Quasi-Linear Systems of PDE
I We make the ansatz
n
! n
!
X Y
u(x) = exp i k xk .U= exp(ik xk ) . U ,
k=0 k=0

x Rn , where 0 , . . . , n R, U Cp .
I Solutions of this type are called plane waves.
I The ansatz leads to
n
! n n
!
X X X
i0 exp i k xk . U + Ak ik exp i k xk . U
k=0 k=1 k=0
n
! n
!
X X
= i exp i k xk k .Ak + 0 .E U=0
k=0 k=1

for every x Rn .
First-Order Quasi-Linear Systems of PDE
I The latter is satisfied if and only if
n
!
X
k .Ak + 0 .E U = 0 (6.22)
k=1

I and hence allows non-trivial U if and only if


n
! n
!
X X
det k .Ak + 0 .E = det k .Ak (0 ).E = 0 .30
k=1 k=1
(6.23)
I Note that in this case 0 , . . . , n , if non-trivial, is normal to
a characteristic hypersurface.

30
In applications, such equation is called a dispersion relation. Such
relation restricts the possible wave vectors = (0 , . . . , n ) Rn+1 .
First-Order Quasi-Linear Systems of PDE
I For arbitrarily given 1 , . . . , n R, the condition (6.23)
states that 0 is an eigenvalue of the matrix

1 .A1 + + n .An (6.24)


I and that U is the corresponding eigenvector.
I We say that (6.21) is hyperbolic if this matrix has only real
eigenvalues such that the associated eigenvectors form a
basis of Cp .
I We say that (6.21) is elliptic if this matrix has no real
eigenvalues.
I If (6.21) is hyperbolic,
I 1 , . . . , n R, 01 , . . . , 0r , where r {1, . . . , p}, are
eigenvalues of the corresponding matrix (6.24) with
corresponding eigenvectors Ul1 , . . . , Ulsl for every
l {1, . . . , r},
First-Order Quasi-Linear Systems of PDE
I where s1 , . . . , sr {1, . . . , p} are such that
X r
sl = p ,
l=1
and such that U11 , . . . , Ursr form a basis of Cp ,
I then u : Rn+1 C defined by
n
! r " sl
#
X X X
u(x) := exp i k xk exp(i0l x0 ) als Uls
k=1 l=1 s=1
n+1
for every x = (x0 , . . . , xn ) R , where a11 , . . . , arsr C,
is a solution of (6.21).
I Moreover, it is to be expected31 that the span of such
solutions contains solutions of (6.21) corresponding to
data, e.g., with compact support, on surfaces of constant
time.
31
For precise analysis, the methods of Fourier analysis need to be used
which are beyond the scope of this course.
First-Order Quasi-Linear Systems of PDE

More generally, for the quasilinear system (6.1), we define


define hyperbolicity and ellipticity by localizing the previous
conditions.
First-Order Quasi-Linear Systems of PDE
Definition 6.4
If A0 = E, we say that the system (6.1) is hyperbolic (elliptic) if
for every x , y 0 , where 0 Rp is an appropriate
subset32 , and (1 , . . . , n ) Rn \ {0}
n
X
k .Ak (x, y)
k=1

has only real eigenvalues with corresponding eigenvectors


forming a basis of Rp (has no real eigenvalues). Sufficient for
this is that all A1 , . . . , An assume values in the linear subspace of
symmetric matrices of M(p p, R). In this case (6.1) is called a
symmetric hyperbolic system. Note that every first-order
quasi-linear PDE for one real-valued function is hyperbolic
according to this definition.
32
For instance, for the system from gas dynamics studied in the subsequent
section, such 0 is given by R2 for > 2 and R R for < 2.
First-Order Quasi-Linear Systems of PDE

Problem 6.5
The source-free Maxwells equations for the electromagnetic
field E, B : R4 R3 in Minkowski space and inertial
coordinates t, x, y, z are given by
E B
= cB , = c E ,
t t
E =B=0 ,

where c denotes the speed of light, and denotes the gradient


in the spacial coordinates x, y, z. Show that the evolution
equations, i.e., the equations containing time derivatives, form a
symmetric hyperbolic system. In addition, find all plane wave
solutions for the whole system.
First-Order Quasi-Linear Systems of PDE

Remark 6.5.1
If (6.1) is elliptic, there are no characteristic hypersurfaces for
any of its solutions.
Remark 6.5.2
For sufficient conditions for the well-posedness of initial value
problems for linear and quasi-linear Hermitian hyperbolic
systems see, e.g., Theorem 6.30, Corollary 6.31, Theorem 11.8,
Remark 11.9 and Theorem 13.6 in [3].
Example 19
The Cauchy-Riemann equations form an elliptic system as a
consequence of the results in Example 18.
First-Order Quasi-Linear Systems of PDE

Example 20
I We consider the system (6.11)
       
1 0 u c2 / u 0
+ = (6.25)
0 1 t u y 0

on R2 from Section 2
I that occurred in the discussion of compressible Euler
equations.
First-Order Quasi-Linear Systems of PDE
Example (cont.)
I Then for every (t, x) R2 and 1 R the corresponding
matrix  
1 u(t, x) 1 c2 (t, x)/(t, x)
1 (t, x) 1 u(t, x)
leads on the eigenvalue equation
 
1 u(t, x) 1 c2 (t, x)/(t, x)
0 = det
1 (t, x) 1 u(t, x)
= (1 u(t, x) )2 12 c2 (t, x)
= [1 (u(t, x) + c(t, x)) ] [1 (u(t, x) c(t, x)) ]

I and hence to the eigenvalues

= 1 [u(t, x) c(t, x)] .


First-Order Quasi-Linear Systems of PDE

Example (cont.)
I These eigenvalues are identical if and only if

1 c(t, x) = 0

which because of (6.10) implies that

(t, x) = 0 .

I In these cases, the matrix is symmetric.


I As a consequence, the system (6.11) is hyperbolic.
First-Order Quasi-Linear Systems of PDE
t

1
2

0 x
0 1 2 3 4 5

Figure 20: Characteristic curves connecting the initial data curve


t = 0 with the point (1, 3). Compare Problem 21.
First-Order Quasi-Linear Systems of PDE

Example 21
Decide whether the system
u v u v
2 2 + 3 =0,
x x t t
u v v
4 + =0 (6.26)
x x t
on R2 is hyperbolic.
First-Order Quasi-Linear Systems of PDE

Example (cont.)
Solution: In matrix form, the system is given by
       
1 3 2 2 u 0
+ =
0 1 t 1 4 x v 0

which is equivalent to the system


       
1 0 5 14 u 0
+ = . (6.27)
0 1 t 1 4 x v 0
First-Order Quasi-Linear Systems of PDE

Example (cont.)
The equation
 
51 + 0 141
det = (51 + 0 )(41 + 0 ) + 1412
1 41 + 0
= 02 + 0 1 612 = (0 + 31 )(0 21 ) = 0

has the real solutions 0 = 31 , 0 = 21 that differ for


1 R . As a consequence, the system (6.26) is hyperbolic.
First-Order Quasi-Linear Systems of PDE

Example 22
Calculate a solution to (6.26) such that

u(0, ) = f , v(0, ) = g , (6.28)

where f , g C 1 (R, R).


First-Order Quasi-Linear Systems of PDE
Example (cont.)
Solution: From the solution of Problem 21, it follows that the
constant vector fields of values (1, 3) and (1, 2) are
characteristic vector fields for the equivalent system (6.27). We
define derivatives

D+ := +3 , D := 2
t x t x
in these directions which leads to the identities
1
= (D+ D )
x 5
3 1
= D+ (D+ D ) = (2D+ + 3 D )
t 5 5
First-Order Quasi-Linear Systems of PDE
Example (cont.)
and
u v u v 2
2 2 + 3 = (D+ D ) u
x x t t 5
1 2 1
+ (2D+ + 3 D ) u (D+ D ) v (6D+ + 9D ) v
5 5 5
1
= (4D+ u + D u 8D+ v 7D v) = 0 ,
5
u v v 1
4 + = (D+ D ) u
x x t 5
4 1
(D+ D ) v + (2D+ + 3 D ) v
5 5
1
= (D+ u D u 2D+ v + 7D v) = 0 .
5
First-Order Quasi-Linear Systems of PDE

Example (cont.)
Hence (6.26) is equivalent to the system

4D+ u + D u 8D+ v 7D v = 0
D+ u D u 2D+ v + 7D v = 0 .

Finally, by linear combination of the last two equations, we


arrive at
D+ (u 2v) = 0 , D (u 7v) = 0
which identifies the functions u 2v and u 7v as Riemann
invariants.
First-Order Quasi-Linear Systems of PDE

Example (cont.)
Since (t, x) = (0, x 3t) + t(1, 3), the points (t, x) R2 and
(0, x 3t) are connected by an integral curve of the constant
vector field (1, 3). As a consequence,

u(t, x) 2v(t, x) = f (x 3t) 2g(x 3t) .

Further, since (t, x) = (0, x + 2t) + t(1, 2), (t, x) R2 and


(0, x + 2t) are connected by an integral curve of the constant
vector field (1, 2). As a consequence,

u(t, x) 7v(t, x) = f (x + 2t) 7g(x + 2t) .


First-Order Quasi-Linear Systems of PDE

Example (cont.)
Hence,
1
u(t, x) = [7f (x 3t) 2f (x + 2t) 14g(x 3t)
5
+14g(x + 2t)]
1
v(t, x) = [f (x 3t) f (x + 2t) 2g(x 3t)
5
+7g(x + 2t)] . (6.29)

Indeed, it can be verified that u, v : R2 R defined by the


equations in (6.29) satisfy (6.26) and (6.28).
First-Order Quasi-Linear Systems of PDE

Problem 6.6
Calculate the C 1 -solutions of
a)
u v u
4 6 + =0,
x x t
u v v
3 + =0
x x t
on R2 for initial data

u(0, x) = sin(x) , v(0, x) = cos(x)

for every x R.
Problem (cont.)
b)
u v u v
3 +2 + + =0,
x x y y
u v u v
5 +2 + =0
x x y y

on R2 for initial data

u(x, 0) = sin(x) , v(x, 0) = ex

for every x R.
First-Order Quasi-Linear Systems of PDE

Problem (cont.)
c)
u v u v
3 + 16 +2 + =0,
x x y y
u v u v
7 + 36 +4 +3 =0
x x y y

on R2 for initial data

u(x, 0) = e3x , v(x, 0) = ex

for every x R.
First-Order Quasi-Linear Systems of PDE
Solutions:
a)
1
u(t, x) = [6 sin(x 3t) 6 cos(x 3t) sin(x + 2t)
5
+6 cos(x + 2t)]
1
v(t, x) = [sin(x 3t) cos(x 3t) sin(x + 2t)
5
+6 cos(x + 2t)]
for all t, x R.
b)
u(x, y) = sin(x + y)
1
8 sin(x 2y) + 6ex2y 8 sin(x + y)

v(x, y) =
6
for all x, y R.
First-Order Quasi-Linear Systems of PDE

c)

u(x, y) = 3 e3(xy) 6 e(xy) 2 e3(x2y) + 6 e2yx


v(x, y) = e3(xy) 2 e(xy) e3(x2y) + 3 e2yx

for all x, y R.
First-Order Quasi-Linear Systems of PDE

For this, let f C 1 (R2 , R). In the following, we calculate the


solutions u C 2 (R2 , R) of the wave equation

2u 2u
2 =f (6.30)
t 2 x
to the initial values
u
u(0, ) = g1 , (0, ) = g2 , (6.31)
t
where g1 C 2 (R, R), g2 C 1 (R, R).
First-Order Quasi-Linear Systems of PDE
I First, we notice the equivalence of (6.30) to a symmetric
hyperbolic system of PDE.
I For this, let u C 2 (R2 , R) be a solution of (6.30).
I Then
2u 2u
   
u u u u
f = 2 2 = + +
t x t t x x t x
v v
= ,
t x
I where v C 1 (R2 , R) is defined by
u u
v := + .
t x
I Hence (u, v) C 2 (R2 , R) C 1 (R2 , R) satisfies the system
u u v v
+ =v, =f . (6.32)
t x t x
I Note that this system is symmetric hyperbolic.
First-Order Quasi-Linear Systems of PDE

I On the other hand if (u, v) C 2 (R2 , R) C 1 (R2 , R) is a


solution of the system (6.32),
I then
   
v v u u u u
f = = + +
t x t t x x t x
2 2
u u
= 2 2
t x
I and hence u is a solution of (6.30).
I In the following, we find the solutions of (6.32) satisfying

u(0, ) = g1 , v(0, ) = g2 + g10 .


First-Order Quasi-Linear Systems of PDE
I For this, let (t, x) R2 .
I Then the characteristic path : R R2 of the second
equation in (6.32) satisfying

(t) = (t, x)

I is given by
( ) = (, x + t )
for every R.
I Hence
Z t
v(t, x) = (g2 + g10 )(x + t) + f (, x + t ) d (6.33)
0

for all (t, x) R2 .


First-Order Quasi-Linear Systems of PDE
I Indeed, v : R2 R defined by (6.33) is an element of
C 1 (R2 , R) satisfying the second equation in (6.32) as well
as v(0, ) = g2 + g10 .
I The characteristic path + : R R2 of the first equation in
(6.32) satisfying
+ (t) = (t, x)
is given by
+ ( ) = (, x t + )
for every R.
I Hence
Z t
u(t, x) = g1 (x t) + v(, x t + ) d (6.34)
0

for all (t, x) R2 .


First-Order Quasi-Linear Systems of PDE
I Indeed, u : R2 R defined by (6.34) is an element of
C 1 (R2 , R) satisfying the first equation in (6.32)
I as well as u(0, ) = g1 .
I The following calculation leads to the representation (6.35)
that shows that indeed u C 2 (R2 , R).
Z t
u(t, x) = g1 (x t) + (g2 + g10 )(x t + 2 )
0
Z 
+ f (s, x t + 2 s) ds d
0
 Z x+t 
1
= g1 (x + t) + g1 (x t) + g2 ( ) d
2 xt
Z t Z t 
+ f (s, x t + 2 s) d ds
0 s
First-Order Quasi-Linear Systems of PDE

I and finally
 Z x+t
1
u(t, x) = g1 (x + t) + g1 (x t) + g2 ( ) d
2 xt
Z t  Z x+ts  
+ f (s, s) ds ds (6.35)
0 x(ts)

for all (t, x) R2 .


I Hence we proved the following
First-Order Quasi-Linear Systems of PDE

Theorem 6.7
Let f C 1 (R2 , R), g1 C 2 (R, R) and g2 C 1 (R, R). Then
there is a unique u C 2 (R2 , R) satisfying the wave equation
(6.30) with initial values (6.31). That u is given by (6.35).
First-Order Quasi-Linear Systems of PDE

Figure 21: Graph of the solution u of (6.30) with vanishing


inhomogeneity and corresponding to the initial data
u(0, x) = 1/(1 + x 2 ) and (u/t)(0, x) = 0 for all x R.
First-Order Quasi-Linear Systems of PDE

Figure 22: Graph of the solution u of (6.30) with vanishing


inhomogeneity and corresponding to the initial data u(0, x) = 0 and
(u/t)(0, x) = 1/(1 + x 2 ) for all x R.
First-Order Quasi-Linear Systems of PDE

Figure 23: Graph of the solution u of (6.30) with inhomogeneity


f (t, x) := 1/(1 + x 2 ) for all (t, x) R2 and corresponding to the
initial data u(0, x) = 0 and (u/t)(0, x) = 0 for all x R.
First-Order Quasi-Linear Systems of PDE

Remark 6.7.1

(i) (Domain of dependence of solutions) According to


(6.35), the value of the solution u at a point (t, x) is
influenced only by the values of g1 at the points x t and
x + t, the values of g2 in the closed interval between
x t, x + t and the values of f inside and on the triangle
with corners (0, x t), (0, x + t), (t, x). Compare Fig 24.
In particular, this allows the calculation of the solution at
previous times from data given at a later time.
Remark (cont.)
(ii) (Conservation of compact support) In the case of
vanishing f and data g1 and g2 vanishing for distances
larger than some L > 0 from the origin, u(t, ) vanishes for
distances larger than |t| + L from the origin because

|x| > |t| + L

for (t, x) R2 implies that

|x + s| > |x| |s| > |x| |t| > L

for all s R such that |s| 6 |t|. Hence the solution u


corresponding to such data is non-zero only on a bounded
domain at every time t R which is typical for hyperbolic
problems. In particular, the outer boundary of the domain
of influence of such data proceeds at speed 1.
Remark (cont.)
(iii) (Well-posedness of the initial value problem) The
solution u is uniquely determined by the data g1 and g2 and
depends continuously on the data in the sense that small
changes of g1 and g2 result only in small changes of u. In
such a case, we say the initial value problem is well-posed.
Remark (cont.)
(iv) (Weak solutions) For the existence of the integrals in the
solution (6.35), it would be sufficient, e.g., that
g1 , g2 C(R, R). This suggests that in such cases the u
defined by (6.35) solves the wave equation where its
derivatives are interpreted in some weak sense.
First-Order Quasi-Linear Systems of PDE

Problem 6.8
Solve the initial value problem for the 1 + 1 wave equation for
reflecting boundary conditions.
First-Order Quasi-Linear Systems of PDE
t

x
x-t x x+t
Figure 24: The past domain of dependence of value of the solution u
of (6.30) at the point (t 0 , x 0 ). That value depends only on the values of
the inhomogeneity f in and on the boundary of the red domain and the
values of the data u(0, x 0 t 0 ), (u/t)(0, x 0 t 0 ), u(0, x 0 + t 0 ),
(u/t)(0, x 0 + t 0 ) at time 0.
First-Order Quasi-Linear Systems of PDE

x
a-t a b b+t
Figure 25: The future domain of influence of the data on the interval
[a, b] at time 0 on the solution u of (6.30). That part of the data can
only influence the values of u inside and on the boundary of the blue
domain.
First-Order Quasi-Linear Systems of PDE

Figure 26: Graph of u from Example 23 corresponding to (6.36).


First-Order Quasi-Linear Systems of PDE

The following example shows that there are solutions of the


wave equation in one space dimension, that, interpreted as a
path in L p -space that is parametrized by time, grow indefinitely.
This result can be considered as already indicated by the fact
that u : R R, defined by

u(t, x) := t

for every (t, x) R2 , is a solution of this equation.


First-Order Quasi-Linear Systems of PDE

Example 23
(Existence of solutions of the wave equation in one space
dimension whose spatial L p -norm is growing indefinitely
with time) In the following, for every p > 0, we show the
existence of a solution u C 2 (R2 , R) of (6.30) for which there
is t0 > 0 and C > 0 such that
Z 1/p
p
|u(t, x)| dx > C t 1/p

for every t > t0 .


First-Order Quasi-Linear Systems of PDE
Example (cont.)
I For this, let p > 0.
I In addition, let C 1 (R, R) be symmetric,
positive-valued, satisfying (0) > 0 and such that
(x) = 0 for all x R \ (1, 1).
I For instance, : R R defined by
(
(1 x 2 )2 if |x| < 1
(x) = (6.36)
0 if |x| > 1

satisfies these conditions.


I Further, let C 2 (R, R) be the antiderivative of
satisfying (0) = 0.
First-Order Quasi-Linear Systems of PDE
Example (cont.)
I We note that for every x R
Z x Z 0
(x) = (y) dy (y) dy ,
c c

where c 6 min{x, 0}.


I As a consequence,
( Rx
(y) dy if x > 0
(x) = R 00
x (y) dy if x < 0

for every x R.
First-Order Quasi-Linear Systems of PDE

Example (cont.)
I In particular, it follows that is monotonically increasing
and antisymmetric.
I Then u C 2 (R2 , R) defined by
1
u(t, x) := [(x + t) (x t)]
2
for every (t, x) R2 is a solution of (6.30)
I such that
u
u(0, x) = 0 , (0, x) = (x)
t
for every x R.
First-Order Quasi-Linear Systems of PDE
Example (cont.)
I Also, u(t, x) = 0 for x > |t| + 1 as well as x 6 (|t| + 1)
and
1 1
u(t, x) = [(x+t)(xt)] > [(x+t)(x+t)] = 0
2 2
for all (t, x) [0, ) R.
I Further, for t > 0 and x R, it follows that
Z x+t Z 0 Z xt
1
u(t, x) := (y) dy (y) dy (y) dy
2 c c c
1 x+t
Z 0  Z
+ (y) dy = (y) dy ,
c 2 xt

where c 6 min{x t, 0}.


First-Order Quasi-Linear Systems of PDE

Example (cont.)
I In particular, if t > 2 and |x| 6 t/2, we conclude that
Z x+t Z t/2
1 1
u(t, x) = (y) dy > (y) dy
2 xt 2 t/2
Z 1
1
> (y) dy > 0
2 1

I and hence that


Z Z t/2  Z 1 p
p p 1
|u(t, x)| dx > |u(t, x)| dx > (y) dy t.
t/2 2 1
Single Second-Order PDE
In the following, we consider second-order quasi-linear
second-order PDE of the form
n
X 2u
aij (x, u(x), (u)(x)) (x)
i,j=0
xi xj
n
X u
+ bi (x, u(x), (u)(x)) (x)
i=0
xi
+ c(x, u(x), (u)(x)) u(x) = f (x, u(x), (u)(x)) (7.1)

for every x , where n N \ {0, 1}, is a non-trivial open


subset of Rn , aij : Rn+2 R and bi : Rn+2 R for
i, j {0, . . . , n}, c : Rn+2 R, f : Rn+2 R and
u C 2 (, R).
Single Second-Order PDE

As a consequence of Schwarzs Theorem, we can assume


without restriction that
aij = aji
for all i, j {0, . . . , n}.
Single Second-Order PDE

I For the definition of characteristic hypersurfaces for (7.1)


and u, we define aij : R, bi : R, c : R and
f : R by

aij (x) := aij (x, u(x), (u)(x)) , bi (x) := bi (x, u(x), (u)(x)) ,
c(x) := c(x, u(x), (u)(x)) , f (x) := f (x, u(x), (u)(x))

for every x and i, j {0, . . . , n}.


I Further, let g C 2 (, R) such that (g)(x) 6= 0 for all
x S, where

S := {x : g(x) = 0} .
Single Second-Order PDE

I Since g C 1 (, Rn+1 ), there is an open neighborhood U


of S in such that

v(x) := (g)(x) 6= 0

for all x U.
I In particular, S is a regular C 2 -hypersurface contained in
with normal field v|S .
I We assume that the restrictions

u u
u|S , ... , (7.2)
x0 S xn S

are given.
Single Second-Order PDE

I Since u C 2 (, R), this implies that the partial derivative


of u on S in any direction of Rn+1 is determined by (7.2).
I The same is true for all partial derivatives of
u u
, ... ,
x0 xn

on S in directions of Rn+1 which are tangential to S.


Single Second-Order PDE

Definition 7.1
The last statement is true for all directions of Rn+1 if and only if
(7.1) can be solved on S for the second derivative
( )( u) into the normal direction v of S.

In this case, we say that S is non-characteristic for u.

If this is nowhere possible on S, we say that S is characteristic


for u.

We extend the previous definitions to all solutions of (7.1).


Single Second-Order PDE
In the following, we investigate the solubility of (7.1) for the
second derivative ( )( u) into the normal direction .
I In a first step, it follows that
n n
2 u
  
X X i u
aij = aij ei + 2
i,j=0
xi xj U i,j=0
|| xj
n n
X u X u
= i aij ( ) + aij (ei ) ,
i,j=0
xj i,j=0
x j

I where
i j
ei := ei = ei i , i := 2
|| 2 ||

for every i {0, . . . , n}.


Single Second-Order PDE

I Further,
u
( ) = ( u) u
xj xj xj

= [(ej + j ) ] ( u) u
xj

= j ( )( u) + (ej )( u) u
xj
= j ( )( u) + [(ej )] u + [(ej )u]

u
xj

for every j {0, . . . , n}.


Single Second-Order PDE

I Hence ( )( u) is determined on S by (7.1) as well


as (7.2) if
n n
X 1 X
aij (x) i (x) j (x) = aij (x) i (x) j (x) 6= 0
i,j=0
|(x)|4 i,j=0

for all x S.
I As a consequence, we make the following definition.
Single Second-Order PDE
Definition 7.2
(i) S = g1 ({0}) is non-characteristic for u if and only if
n
X g g
aij (x, u(x), (u)(x)) (x) (x) 6= 0
i,j=0
xi xj

for all x S.
(ii) S = g1 ({0}) is characteristic for u if and only if
n
X g g
aij (x, u(x), (u)(x)) (x) (x) = 0
i,j=0
xi xj

for all x S.
Single Second-Order PDE

Figure 27: Two characteristic planes of (7.3) through the origin in R3 .


Here t := x0 , x := x1 , y := x2 . Compare Problem 24.
Single Second-Order PDE

Figure 28: Forward and backward characteristic cone in R3 with apex


at the origin, where t := x0 , x := x1 , y := x2 . Compare Problem 24.
Single Second-Order PDE
Example 24
Find characteristic surfaces through any point
(y0 , . . . , yn ) Rn+1 of the equation

2u
4u + c u = 0 , (7.3)
x02

where n N , c R, u C 2 (Rn+1 , R) and 4 is the Laplace


operator in n-dimensions. The action of the last on every twice
partially differentiable function f defined on some non-empty
open subset of Rn+1 is defined by
n
X 2f
4f := .
k=1
xk2
Single Second-Order PDE
Example (cont.)
Solution:
I For this, let k Rn+1 \ {0} and gk C (Rn+1 , R) be
defined by
n
X
gk,y (x) := ki (xi yi )
k=0

for every x Rn+1 .


I Then
Sk,y := {x Rn+1 : gk (x) = 0}
is a plane through y with normal field

= (gk )|Sk,y = k .
Single Second-Order PDE

Example (cont.)
I Hence Sk,y is characteristic if and only if k is a null vector
in the Minkowski metric for Rn+1 ,
I i.e., if
X n
2
k0 kj2 = 0 .
j=1
Single Second-Order PDE
Example (cont.)
I Note that also

Cy := {x Rn+1 : g(x) = 0} ,

I where gy C (Rn+1 \ {y}, R) is defined by


n
X
2
gy (x) := (x0 y0 ) (xj yj )2
j=1

for every x Rn+1 , is characteristic.


I It is the union of the backward characteristic cone Cy with
apex at y and the forward characteristic cone Cy+ with apex
at y.
Single Second-Order PDE

Example (cont.)
I The latter are defined by

Cy := {x Rn+1 : x0 6 y0 , gy (x) = 0} ,
Cy+ := {x Rn+1 : x0 > y0 , gy (x) = 0} .

We define A C 2 ( Rn+2 , M((n + 1) (n + 1), R)) by

A := (aij )i,j{0,...,n} .
Single Second-Order PDE
Definition 7.3
We say that the equation (7.1) is of
(i) parabolic type at x if at least one of eigenvalues of
A(x, y) vanishes for every y Rn+2 .
(ii) elliptic type at x if A(x, y) has only non-vanishing
eigenvalues of the same sign for every y Rn+2 .
(iii) hyperbolic type at x if for every y Rn+2 the
corresponding A(x, y) has only non-vanishing eigenvalues
of which n 1 have the same sign, whereas the remaining
eigenvalue is of opposite sign.
(iv) ultra-hyperbolic type at x if A(x, y) has only
non-vanishing eigenvalues with at least two eigenvalues of
each sign for every y Rn+2 . Note that this implies that
n > 4.
Single Second-Order PDE

Remark 7.3.1
If (7.1) is elliptic there are no characteristic hypersurfaces for
any of its solutions.
Example 25
The velocity potential of a stationary, isentropic, irrotational,
two-dimensional flow (u, v) of a compressible ideal fluid is
described by

2 2 2
2
(c2 u2 ) 2uv + (c 2
v ) =0, (7.4)
x 2 xy y2
where c denotes the local speed of sound.
Single Second-Order PDE
Example (cont.)
The solutions of
2
c u 2 uv
0=
uv c2 v2
= (c2 u2 )(c2 v2 ) u2 v2
= 2 (2c2 u2 v2 ) + (c2 u2 )(c2 v2 ) u2 v2
= 2 (2c2 u2 v2 ) + c2 (c2 u2 v2 )
2
u2 + v 2 (u2 + v2 )2
 
2
= c
2 4

are given by
{c2 , c2 (u2 + v2 )} .
Single Second-Order PDE
Example (cont.)
Hence (7.4) is elliptic at points (x, y) of subsonic speed

u2 (x, y) + v2 (x, y) < c2 (x, y) ,

parabolic at points (x, y) of sonic speed

u2 (x, y) + v2 (x, y) = c2 (x, y) ,

and hyperbolic at points (x, y) of supersonic speed

u2 (x, y) + v2 (x, y) > c2 (x, y) .

Such PDE as (7.4) that have different type in different parts of


the underlying domain are called of mixed type.
Single Second-Order PDE
Example 26
(Tricomis equation) Tricomis equation is given by

2u 2u
y (x, y) + (x, y) = 0 (7.5)
x 2 y2

for (x, y) R2 . It is a model for the transition from subsonic to


supersonic speeds in aerodynamics. For every (x, y) R2 , the
solutions of

y 0
0 = = (y )(1 )
0 1

are given by = 1, y. Hence (7.5) is elliptic in the open upper


half-plane, parabolic on the real axis and hyperbolic in the open
lower half-plane.
Single Second-Order PDE
y

9 7 5 3 1 1 3 5 7 9
x
-2 -2 -2 -2 -2 2 2 2 2 2

-4

Figure 29: Characteristic curves of Tricomis equation (7.5). Compare


Problem 27.
Single Second-Order PDE

Example 27
Calculate two characteristic curves of the Tricomi equation
(7.5) through every (x, y) R2 where possible.
Single Second-Order PDE

Example (cont.)
Solution:
I For this, let be a non-trivial open subset of R2 ,
I g C 2 (, R) such that v(x, y) := (g)(x, y) 6= 0 for all
(x, y) S,
I where
S := {(x, y) : g(x, y) = 0} .
I S is characteristic if and only if

y v12 (x, y) + v22 (x, y) = 0

for all (x, y) S.


Single Second-Order PDE
Example (cont.)
I As a consequence, there are no characteristic curves in
R [0, ).
I Therefore, in the following we assume that
R (, 0).
I Further, the condition that
 2  2
2 2 g g
0 = y v1 (x, y) + v2 (x, y) = y (x, y) + (x, y)
x y

   
g g g g
= (x, y) y (x, y) (x, y) + y (x, y)
y x y x

for all (x, y) S


Single Second-Order PDE

Example (cont.)
I is satisfied if
g g
(x, y) y (x, y) = 0 (7.6)
y x

for all (x, y) .


I The latter equations are single first-order linear PDE that
can be solved by the characteristic method (see Section 2)
or the following educated guess.
Single Second-Order PDE

Example (cont.)
I For the solutions of (7.6) we make the ansatz

g(x, y) = f (F(y) x)

for every (x, y) , where f and F are suitable


C 2 -functions that are to be determined.
Single Second-Order PDE

Example (cont.)
I Then
g g 
(x, y) y (x, y) = f 0 (F(y)x) F 0 (y) y = 0
y x

is satisfied for every f C 2 (R, R) and


F C 2 ((, 0), R) defined by
2
F(y) := (y)3/2
3
for all y < 0.
Single Second-Order PDE

Example (cont.)
I In particular, for every c R, the zero set
(  2/3 )
3
Sc, := (x, y) R2 : y = (x c)
2

of the function gc, : R (, 0)


I defined by
2
gc, (x, y) := (y)3/2 (x c)
3
for all (x, y) R (, 0) is a characteristic curve.
Single Second-Order PDE

Single Second-Order PDE


I Through a given point (x0 , y0 ) R (, 0),
characteristics are given by Sc1 ,+ , Sc2 , , where

2 2
c1 := x0 (y0 )3/2 , c2 := x0 + (y0 )3/2 .
3 3

Problem 7.4
In [23]: Problems 2.1 a), d) and 2.2 of Chapter V, Section 2 and
7.7 a), b) and c) of Chapter V, Section 7.
Single Second-Order PDE

In the following, we investigate further the case of linear


second-order PDE with constant coefficients, i.e., the special
case of (7.1), where n N , = Rn+1 and that aij R, bi R,
c R are constant for all i, j {0, . . . , n}.
Single Second-Order PDE

I In the first step, we try to simplify the structure of (7.1)


by a linear bijective coordinate transformation
h : Rn+1 Rn+1 ( C (Rn+1 , Rn+1 )) adapted to its
coefficients.
I We define
v := u h1
such that
u=vh .
Single Second-Order PDE
I Then it follows by the chain rule that
n
u X hl v
(x) = (x) (h(x)) ,
xj l=0
xj xl
n
2u X 2 hl v
(x) = (x) (h(x))
xi xj l=0
xi xj xl
n
X hl hk 2v
+ (x) (x) (h(x))
k,l=0
xj xi xk xl
n n
X hk hl 2v X 2 hk v
= (x) (x) (h(x)) + (x) (h(x))
k,l=0
xi xj xk xl
k=0
x i xj xk

for every x Rn+1 and i, j {0, . . . , n}.


Single Second-Order PDE
I Since h is assumed linear, all partial derivatives of the first
order are constant.
I Hence it follows
n
u X v
(x) = Hki (h(x)) ,
xi k=0
xk
n
2u X 2v
(x) = Hki Hlj (h(x))
xi xj k,l=0
xk xl

for every x Rn+1 and i, j {0, . . . , n},


I where
hi
Hij := .
xj
Single Second-Order PDE

I Substitution into (7.1) leads to the equivalent equation


n n n n
X X 2v X X v
aij Hki Hlj + bi Hki +cv
i,j=0 k,l=0
xk xl i=0 k=0
xk

(7.7)
n n
! n n
!
X X 2v X X v
= Hki aij Hlj + Hki bi +cv
k,l=0 i,j=0
xk xl k=0 i=0
xk
1
=f h

for v.
Single Second-Order PDE
I We define the matrices33

H := (Hij )i,j{0,...,n}

I and !
n
X
A := Hki aij Hlj .
i,j=0 k,l{0,...,n}

I Then it follows that

A = H A H t ,

where denotes matrix multiplication and H t the transpose


of H.
33
Note that H is the representation of the derivative of H with respect to the
canonical basis of Rn+1 .
Single Second-Order PDE

I According to Linear Algebra, there exists an orthogonal H


such that
H A H t = diag(1 , . . . , n ) ,
I where 1 , . . . , n denote the eigenvalues of A and
diag(1 , . . . , n ) the diagonal matrix containing these
eigenvalues in the diagonal.
Single Second-Order PDE

I For such H, (7.7) simplifies to


n n
X 2 v X v
k + bk + c v = f h1
k=0
xk2 k=0 xk

which is called the canonical form of (7.1) in the case of


constant coefficients.
I Here n
X
bk := Hki bi
i=0

for k {1, . . . , n}
Single Second-Order PDE
I In the cases of non-vanishing eigenvalues 1 , . . . , n , a
further scale transformation leads to the equivalent
equation for w
n n
X k 2 w X bk w
2
+ + c w = f h1 s1 ,
k=0
|k | xk k=0
|k |1/2 x
k

I where
w := v s1 ,
and s : Rn+1 Rn+1 is defined by

s(x0 , . . . , xn ) := |1 |1/2 x0 , . . . , |n |1/2 xn




for all (x0 , . . . , xn ) Rn+1 .


Single Second-Order PDE
I Finally, in this case further simplification is possible by the
introduction of
n
!
1 X bk
w(x) := exp xk w(x)
2 k=0 |k |3/2

for all (x0 , . . . , xn ) Rn+1


I which leads to the equivalent equation for w
" n n
! #
X k 2 w 1 X bk2
2
+ c w (x)
k=0
| k | xk 4 | k|
n
! k=0
1 X k bk
xk f h1 s1 (x)

= exp
2 k=0 |k | 3/2

for all (x0 , . . . , xn ) Rn+1 .


Single Second-Order PDE
I Hence in the elliptic and hyperbolic cases, we can restrict
the discussion of equations of the form (7.1) in the case of
constant coefficients to two equations.
I In the elliptic case to
4u + c u = f ,
where 4 denotes the Laplace operator in n + 1 dimensions
I and in the hyperbolic case to
2u
4u + c u = f , (7.8)
x02
where 4 denotes the Laplace operator in n dimensions.
I Also we identify the heat equation
T
4T = 0 (7.9)
t
as a parabolic equation.
Single Second-Order PDE

In the following, we calculate the solutions u C 2 (R4 , R) of


the wave equation
2u
4u = 0 (7.10)
t 2
to the initial values
u
u(0, ) = f1 , (0, ) = f2 , (7.11)
t
where f1 C 2 (R3 , R), f2 C 1 (R3 , R).
Single Second-Order PDE

I This is achieved by help of an integral transformation I that


averages the solutions over spheres of radius r.
I From the averaged functions, the solutions are recovered in
the limit r 0.
I As a function of t and r, rescaled averaged functions u
satisfy the wave equation in one space dimension (6.30).
I The associated initial value problem for the functions u is
solved by Theorem 6.7.
Single Second-Order PDE

Figure 30: The domain of integration in the definition of I(t, r, x) is


Sr2 (x), a sphere of radius r around x.
Single Second-Order PDE
I For every h C(R4 , R), we define I(h) : R5 R by
Z
1
I(h)(t, r, x) := h(t, x + ry) dSy
4 S2
Z
1
:= h(t, x + r p(, )) sin dd
4 (0,)(,)

for all (t, r, x) R5 .


I Further, p : R2 R3 is defined by

p(, ) := (sin cos , sin sin , cos )

for all (, ) R2 .
I Then

p((0, ) (, )) = S 2 \ ((, 0] {0} R) .


Single Second-Order PDE
I For (t, r, x) R5 , I(h)(t, r, x) can be interpreted as the
average of h(t, ) over a sphere of radius |r| around x.
I In particular, it follows that
I(h)(t, r, x) = I(h)(t, r, x) , I(h)(t, 0, x) = h(t, x) .
(7.12)
I If in addition h C 2 (R4 , R), then I(h) C 2 (R5 , R).
I In this case, it follows for t R, r > 0, x R3 that
Z
h(t, x + z) dz
Ur (0)
Z
= h(t, x + rp(, )) r 2 sin drdd
(0,r)(0,)(0,2)
Z r
= 4r 2 I(h)(t, r, x) dr ,
0
where Ur (0) denotes the open unit ball around the origin in
R3 .
Single Second-Order PDE

I Hence
 Z r  Z
2 1
4 r I(h)(t, r, ) dr (x) = (4h)(t, x + z) dz
0 4 Ur (0)
Z
1
= (h)(t, x + z) dSz
4 S2 (r)
Z
1
= (h)(t, x + rp(, )) sin() r 2 p(, ) dd
4 (0,)(,)
 
2
=r I(h) (t, r, x) .
r
Single Second-Order PDE

I Since
 Z 

4 r I(h)(t, r, ) dr (r, x) = r 2 [4I(h)](t, r, x) ,
2
r 0

I it follows that
   2 
2 2
0= r r 4 I(h) = r 4 rI(h) ,
r r r 2

where r denotes the coordinate projection onto the second


coordinate of R5 .
I Note that the last equation is valid on the whole of R5 as a
consequence of the first identity in (7.12).
Single Second-Order PDE
I Hence it follows that
 2 

4 rI(h) = 0 . (7.13)
r 2

I In the following, we define for every h C(R4 , R) the


function I(h) : R5 R by

I(h)(t, r, x) := rI(h)(t, r, x)

for every (t, r, x) R5 .


I Furthermore, if h C 2 (R4 , R), then it follows from (7.13)
that
2
I(4h) = 4I(h) = 2 I(h) .
r
Single Second-Order PDE
I Now let f1 C 2 (R3 , R), f2 C 1 (R3 , R)
I and u C 2 (R4 , R) be a solution of the initial value
problem (7.10), (7.11).
I Then
2
 2 
u 2
I(u) = 4 I(u) = I(4u) = I = I(u) .
r 2 t 2 t 2
I In addition,
Z Z
r r
I(u)(0, r, x) = f1 (x + ry) dSy = f1 (x ry) dSy ,
4 S2 4 S2
  Z
r
I(u) (0, r, x) = f2 (x + ry) dSy
t 4 S2
Z
r
= f2 (x ry) dSy
4 S2
for all (r, x) R4 .
Single Second-Order PDE

I Hence it follows by Theorem 6.7 that


 Z
1
I(u)(t, r, x) = (t + r) f1 (x + (t + r)y) dSy
8 S2
Z
(t r) f1 (x + (t r)y) dSy
S2
Z t+r  Z  
+ f2 (x + y) dSy d
tr S2

for all (t, r, x) R5


Single Second-Order PDE
I and
 Z
1
I(u)(t, r, x) = (t + r) f1 (x + (t + r)y) dSy
8r 2
Z S
(t r) f1 (x + (t r)y) dSy
S2
Z t+r  Z  
+ f2 (x + y) dSy d
tr S2

for all (t, r, x) R R R3 .


I Finally, it follows by the second identity in (7.12) that
Z
1
u(t, x) = [ f1 + t (f2 + y f1 ) ] (x + ty) dSy (7.14)
4 S2

for all (t, x) R4 .


Single Second-Order PDE

Theorem 7.5
(Kirchhoffs formula) Let f1 C 3 (R3 , R), f2 C 2 (R3 , R).
Then there is a unique u C 2 (R4 , R) satisfying (7.10) with
initial values (7.11). That u is given by (7.14). Also
  Z 
4 t
u(t, x) = R R, (t, x) 7 f1 (x + ty) dSy (t, x)
t 4 S2
Z
t
+ f2 (x + ty) dSy (7.15)
4 S2

for all (t, x) R4 .


Single Second-Order PDE

Proof.
I In a first step, it will be proved that u2 : R4 R defined by
Z
t
u2 (t, x) := f2 (x + ty) dSy
4 S2

for all (t, x) R4


I is a C 2 -solution of (7.10) such that
u2
u2 (0, ) = 0 , (0, ) = f2 . (7.16)
t
I Indeed, it follows by differentiation under the integral sign
that u2 C 2 (R4 , R).
Single Second-Order PDE

Proof (cont.).
I Also, the first identity in (7.16) is trivially satisfied.
I Further, it follows for (t, x) R4 that
Z Z
u2 1 t
(t, x) = f2 (x + ty) dSy + (y f2 )(x + ty) dSy ,
t 4 S2 4 S2
(7.17)

I and hence that the second identity in (7.16) is satisfied, too.


Single Second-Order PDE
Proof (cont.).
I In addition, if t > 0, it follows by Gauss Theorem
Z
(y f2 )(x + ty) dSy
S 2Z

= (p(, ) f2 )(x + tp(, )) sin dd


S 2Z
1
= 2 (4f2 )(x + z) dz
t Ut (0)
Z
1
= 2 (4f2 )(x + p(, )) 2 sin() d dd
t (0,t)(0,)(,)
1 t 2
Z Z 
= 2 (4f2 )(x + y)dSy d .
t 0 S2
Single Second-Order PDE

Proof (cont.).
I Hence it follows that
2 u2
Z
1
(t, x) = (y f2 )(x + ty) dSy
t 2 2 S2
  Z
t 2
+ (y f2 )(x + ty) dSy
4 t S2
Z 
1 2
+ 2t (4f2 )(x + ty) dSy
t S2
Z
t
= (4f2 )(x + ty) dSy = (4u2 )(t, x)
4 S2
Single Second-Order PDE
Proof (cont.).
I Since 2 u2 /t 2 and 4u2 are continuous and satisfy

2 u2 2 u2
(t, x) = (t, x) , (4u2 )(t, x) = (4u2 )(t, x)
t 2 t 2
for all (t, x) R4 ,
I it follows that u satisfies (7.10).
I In the next step, we define v1 : R4 R by
Z
t
v1 (t, x) := f1 (x + ty) dSy
4 S2

for all (t, x) R4 .


Single Second-Order PDE

Proof (cont.).
I Then it follows by differentiation under the integral sign
that v1 C 3 (R4 , R).
I Also, the previous step implies that

2 v1
4v1 = 0 .
t 2
I We define u1 C 2 (R4 , R) by

v1
u1 := .
t
Single Second-Order PDE
Proof (cont.).
I Then u1 satisfies (7.10) and
v1
u1 (0, ) = (0, ) = f1 ,
t
u1 2 v1
(0, ) = (0, ) = 4v1 (0, ) = 0 .
t t 2
I Hence u := u1 + u2 C 2 (R4 , R) satisfies (7.10) and (6.31).
I That u coincides with (7.14), (7.15) follows with the help
of (7.17).
I The uniqueness of u is a consequence of the reasoning
leading to (7.14).
Single Second-Order PDE

Remark 7.5.1
(i) (Domain of dependence, Huygens principle) According
to (7.15), the value of u at (t, x) R4 depends only the
values of the data on the sphere of radius |t| around x in R3 .
This phenomenon is known as Huygens principle. It is
known to be true for the wave equation in odd space
dimensions different from 1.34

34
In general, it is not valid in the case of a non-vanishing function c in (7.8).
Single Second-Order PDE

Remark (cont.)
(i) Hence for t > 0 the domain of dependence of u(t, x) is the
intersection of backward characteristic cone with apex at
(t, x) with the data surface {0} R3 . For t < 0 it is the
intersection of the forward characteristic cone with apex at
(t, x) with the data surface. 35 In particular, this allows the
calculation of the solution at previous times from the data
at later time which is typical for hyperbolic equations.

35
Note that these cones are both characteristic surfaces. In particular, note
the similarity of the situation to the case of a partial differential equation for
one real-valued function.
Single Second-Order PDE

Remark (cont.)
(ii) (Conservation of compact support) If the data f1 and f2
vanish for distances larger than some radius of R 6 0 from
the origin the u(t, ) vanishes for distances larger than the
radius of |t| + R from the origin because

|x| > |t| + R

for (t, x) R4 implies that

|x + ty| > |x| |t| |y| = |x| |t| > R

for all y S 2 .
Single Second-Order PDE

Remark (cont.)
(ii) Hence the solution u corresponding to such data is
non-zero only on a bounded domain at every time t R
which is typical for hyperbolic problems. In particular, the
outer boundary of the domain of influence of such data
proceeds at speed 1. Such behavior of the solutions is
crucial for the compatibility of PDE / Systems of PDE with
Einsteins Special Relativity because the fastest signal
speed permitted by this theory is the speed of light in
vacuum c = 2.99792458 108 m/sec [5].
Single Second-Order PDE

Remark (cont.)
(iii) (Well-posedness of the initial value problem) The
solution u is uniquely determined by the data f1 and f2 and
depends continuously on the data in the sense that small
changes of f1 and f2 will result only in small changes of u,
i.e., the initial value problem for the wave equation in three
space dimensions for data given on surfaces of constant
time is well-posed.
Single Second-Order PDE

Remark (cont.)
(iv) (Weak solutions) For the existence of the integrals in the
representation in the formula (7.14), it is sufficient, e.g.,
that f1 C 1 (R3 , R) and f2 C(R3 , R). This suggests that
in such cases the u defined by (7.14) solves the wave
equation where its derivatives are interpreted in some weak
sense. Indeed, the adoption of such interpretation is crucial
to achieve a formulation where the order of
differentiability of the restrictions of the solutions to
surfaces of constant time is preserved by the evolution.
Single Second-Order PDE

I For the case of the wave equation in one space dimension,


the solution u is as differentiable as the initial data,
I i.e., to data u(0, ) C 2 (R, R) and
(u/t)(0, ) C 1 (R, R)
I there corresponds a solution u C 2 (R2 , R).
I Theorem 7.5 does not make such a statement.
I Indeed, we will see in the following that such a statement
would be false in general due to a focusing effect.
I For space dimensions n > 1, irregularities in the initial
data are focused from different localities into a smaller
set called caustic leading to stronger irregularities.
Single Second-Order PDE

I For demonstration, we assume in addition that f1 and f2 are


spherically symmetric, i.e., f1 = f | |2 and f2 = g | |2 for
some f C k+1 (R, R), g C k (R, R) and some k N .
I Then,
Z Z
t t
f2 (ty) dSy = f2 (t p(, )) sin dd
4 S2 4 (0,)(,)
Z
t
= g(t 2 ) sin dd = t g(t 2 )
4 (0,)(,)

for every t R.
Single Second-Order PDE

I Hence (7.14), (7.15) give for the corresponding value of u


in the origin

u(t, 0) = f (t 2 ) + 2t 2 f 0 (t 2 ) + tg(t 2 )

for every t R and u(, 0) C k (R, R).


I Therefore in general, u C k (R4 , R), only.
Single Second-Order PDE

In the following, we calculate a solution u C 2 (R3 , R) of the


wave equation
2u
4u = 0 (7.18)
t 2
to the initial values
u
u(0, ) = f1 , (0, ) = f2 , (7.19)
t
where f1 C 3 (R2 , R), f2 C 2 (R2 , R).
Single Second-Order PDE
I This is done by descending from solutions of the wave
equation in three space dimensions,
I i.e., we extend f1 and f2 to functions f1 C 3 (R3 , R) and
f2 C 2 (R3 , R) by
f1 (x1 , x2 , x3 ) := f1 (x1 , x2 ) , f2 (x1 , x2 , x2 ) := f2 (x1 , x2 )

for all x R3 .
I Then the solution u C 2 (R3 , R) of the wave equation in
three dimensions corresponding to data f1 , f2 at time zero is
given by
  Z 
4 t
u(t, x) = R R, (t, x) 7 f1 (x + ty) dSy (t, x)
t 4 S2
Z
t f2 (x + ty) dSy
+
4 S2
for all (t, x) R4 .
Single Second-Order PDE
I Further,
Z Z
f2 (x + ty) dSy = f2 (x + t p(, )) sin dd
S2 (0,)(,)
Z
= f2 (x1 + t sin cos , x2 + t sin sin ) sin dd
(0,)(,)
Z
=2 f2 (x1 + t sin cos , x2 + t sin sin ) sin dd
(0,/2)(,)
f2 (x1 + t sin cos , x2 + t sin sin )
Z
=2 sin cos dd
(0,/2)(,) cos
f2 (x1 + t y1 , x2 + t y2 )
Z
=2 p dy
U1 (0) 1 |y|2
Single Second-Order PDE
I and hence
" !#
f1 (x + t y)
Z
t
u(t, x) = R4 R, (t, x) 7 p dy (t, x)
t 2 U1 (0) 1 |y|2
f2 (x + t y)
Z
t
+ p dy .
2 U1 (0) 1 |y|2
Single Second-Order PDE

I In particular, u satisfies

u
= 0 , u(0, x1 , x2 , x3 ) = f1 (x1 , x2 , x3 ) = f1 (x1 , x2 ) ,
x3
u
(0, x1 , x2 , x3 ) = f2 (x1 , x2 , x3 ) = f2 (x1 , x2 )
t
for all x R3 .
I Hence u := u(, , , 0) C 2 (R3 , R) satisfies (7.18) and
(7.19).
I As a consequence, we conclude the following theorem:
Single Second-Order PDE

Theorem 7.6
Let f1 C 3 (R2 , R), f2 C 2 (R2 , R). Then there is
u C 2 (R3 , R) satisfying (7.18) with initial values (7.19). Such
u is given by
" Z !#
t f1 (x + t y)
u(t, x) := R3 R, (t, x) 7 p dy (t, x)
t 2 U1 (0) 1 |y|2
Z
t f2 (x + t y)
+ p dy (7.20)
2 U1 (0) 1 |y|2

for all (t, x) R3 .


Single Second-Order PDE
Remark 7.6.1
(i) (Domain of dependence) According to (7.20), the value of
u at (t, x) R3 depends only the values of the data on a
circle of radius |t| around x in R2 and its inside. Hence
Huygens principle is not valid for the wave equations in
two space dimensions. Thus for t > 0 the domain of
dependence of u(t, x) is the intersection of the backward
characteristic cone with apex at (t, x) and its inside with
the data surface {0} R2 . For t < 0 it is the intersection
of forward characteristic cone with apex at (t, x) and its
inside with the data surface.36 In particular, this allows the
calculation of the solution at previous times from the data
at a later time.
36
Note that these cones are both characteristic surfaces. In particular, note
the similarity of the situation to the case of a partial differential equation for
one real-valued function.
Single Second-Order PDE
Remark (cont.)
(ii) (Preservation of compact support) If the data f1 and f2
vanish for distances larger than some radius of R 6 0 from
the origin, u(t, ) vanishes for distances larger than the
radius of |t| + R from the origin because

|x| > |t| + R

for (t, x) R3 implies that

|x + ty| > |x| |t| |y| > |x| |t| > R

for all y U1 (0). Hence the solution u corresponding to


such data is non-zero only on a bounded domain at every
time t R. In particular, the outer boundary of the domain
of influence of such data proceeds at speed 1.
Single Second-Order PDE

Remark (cont.)
(iii) (Well-posedness of the initial value problem) The
solution u is uniquely37 determined by the data f1 and f2
and depends continuously on the data in the sense that
small changes of f1 and f2 will result only in small changes
of u, i.e., the initial value problem for the wave equation in
two space dimensions for data given on t = constant
surfaces is well-posed.

37
The uniqueness of the solution follows from Theorem 7.14.
Single Second-Order PDE

Remark (cont.)
(iv) (Weak solutions) For the existence of the integrals in the
representation in the formula (7.20), it would be sufficient,
e.g., that f1 C 1 (R2 , R) and f2 C(R2 , R). This suggests
that in such cases the u defined by (7.20) solves the wave
equation where its derivatives are interpreted in some weak
sense.
Single Second-Order PDE

Problem 7.7
Decide whether there is a focusing effect for the solutions of the
wave equation in two space dimensions.
Single Second-Order PDE
Problem 7.8
Let m > 0 and f C 2 (R, R). Show that a solution of the initial
value problem
u
u(0, ) = 0 , (0, ) = f
t
for the Klein-Gordon equation

2 u 2 u
2 + m2 u = 0 (7.21)
t 2 x
and u C 2 (R2 , R) is given by

1 x+t
Z p 
u(t, x) = J0 m t 2 (x y)2 f (y) dy
2 xt

for every (t, x) R2 .


Single Second-Order PDE

Problem (cont.)
Here, the Bessel function J0 is defined by
Z /2
2
J0 (x) := cos(x sin ) d
0

for every x R. [Hint: Descend to (7.21) from (7.18) by


using the ansatz

u(t, x1 , x2 ) = cos(mx2 ) u(t, x1 ) ,

where t, x1 , x2 R.]
Single Second-Order PDE
In the case of one space dimension, the solution of the
inhomogeneous wave equation is given by (6.35).

In the following, let n {2, 3} and f C 2 (Rn+1 , R). We


calculate a special solution u C 2 (Rn+1 , R) of the
inhomogeneous wave equation

2u
4u = f (7.22)
t 2
to the initial values
u
u(0, ) = 0 , (0, ) = 0 . (7.23)
t
Solutions to more general initial data are given by adding this
solution to solutions of the associated homogeneous wave
equation corresponding to those data.
Single Second-Order PDE
I For the solution, we use Duhamels integral (or method
of impulses).
I For R, let U(, , ) C 2 (Rn+1 , R) be the solution of
the associated homogeneous wave equation to the initial
values
U
U(0, , ) = 0 , (0, , ) = f (, ) .
t
I We define Z t
u(t, x) := U(t , x, ) d
0
n+1
for every (t, x) R .
I Then it follows that
Z t Z t
u U U
(t, x) = (t , x, ) d + U(0, x, t) = (t , x
t 0 t 0 t
Z t 2
2u U U
2
(t, x) = 2
(t , x, ) d + (0, x, t)
t 0 t t
Single Second-Order PDE
Theorem 7.9
Let n {2, 3}, f C 2 (Rn+1 , R). Then there is
u C 2 (Rn+1 , R) satisfying (7.22) with the initial values (7.23).
Such u is given by
Z t !
f (, x + (t ) y)
Z
1
u(t, x) = (t ) p dy d
2 0 U1 (0) 1 |y|2
(7.24)

for all (t, x) R3 if n = 2, and


Z t Z 
1
u(t, x) = (t ) f (, x + (t )y)dSy d
4 0 S2
(7.25)

for all (t, x) R4 if n = 3.


Single Second-Order PDE

Remark 7.9.1
According to (7.25), for n = 3, the domain of dependence of the
value of u at (t, x) R4 depends only the values of f on the part
of the characteristic cone with apex at x with time coordinates
between 0 and t and the values of the data on the intersection of
this cone with the data surface.

On the other hand, for n = 2, according to (7.24), the domain of


dependence of the value of u at (t, x) R3 depends on the
values of f on and insight the characteristic cone with apex at x
with time coordinates between 0 and t and the values of the data
on and inside the intersection of this cone with the data surface.
Single Second-Order PDE

In the following, we consider the equation


n  
X 1 u
aij + bu = 0 , (7.26)
i,j=0
xi xj

where n N , is a non-empty open subset of Rn+1 ,


aij C 1 (, R) for all i, j {0, . . . , n} such that aij = aji for all
i, j {0, . . . , n}, C 1 (, R) such that Ran (0, ),
b C(, R) are given and u C 2 (, R) is the unknown
function.
Single Second-Order PDE

I For solutions u, v of (7.26), it can be derived a


corresponding conservation law.
I First, it follows that
n  
X u v
aij + b uv = 0 ,
i,j=0
xi xj
n  
X v u
aij + b vu = 0 .
i,j=0
x i x j
Single Second-Order PDE

I The difference of the last equations gives


n     
X u v v u
0= aij aij
i,j=0
x i x j x i xj
n     
X 1 v u
= u aij v aij
i,j=0
x i x j x i xj
n   
X 1 v u
= aij u v
i,j=0
x i x j xj
n
" n  #
X 1 X v u
= aij u v
i=0
xi j=0
xj xj
Single Second-Order PDE

I and hence
j(u, v) = 0 ,
I where the vector field j(u, v) C 1 (, Rn+1 ) is defined by
n  
X v u
(j(u, v))i = aij u v
j=0
xj xj

for all i {0, . . . , n}.


I As a consequence, we showed the following theorem:
Single Second-Order PDE
Theorem 7.10
Let n N , a non-empty open subset of Rn+1 , aij C 1 (, R)
for all i, j {0, . . . , n} such that aij = aji for all
i, j {0, . . . , n}, C 1 (, R) such that Ran (0, ),
b C(, R) and u, v C 2 (, R) such that
n   n  
X 1 u X 1 v
aij +b u = aij +b v = 0 ,
i,j=0
xi x j i,j=0
xi xj

Then
j(u, v) = 0 ,
where the vector field j(u, v) C 1 (, Rn+1 ) is defined by
n  
X v u
(j(u, v))i = aij u v
j=0
xj xj
Single Second-Order PDE

and the following simple Corollary:


Corollary 7.11
In addition, if k {2, 3, . . . } and F : C k (, R) C 2 (, R) is a
symmetry transformation, i.e., maps C k -solutions of (7.26) into
C 2 -solutions of (7.26), then it follows the constraint

j(u, F(u)) = 0

for all C k -solutions u of (7.26).


Single Second-Order PDE

Example 28
We consider the special case of the equation

2u
4u + Vu = 0 , (7.27)
t 2
where V C(, R) is such that

V
=0.
t
Single Second-Order PDE

Example (cont.)
If u is a C 3 -solution of (7.27), then

2u 2 u
 
u u
0= 2
4u + Vu = 2 4 +V =0,
t t t t t t

and hence u/t is a C 2 -solution of (7.27). As a consequence,


  n  
u X u
j0 u, = jk u, .
t t k=1
xk t
Single Second-Order PDE
Example (cont.)
Further,
 2  2
2u
 
u u u
j0 u, =u 2 = u (4u Vu)
t t t t
"  # "  #
2 2
u u
= u4u + Vu2 = (uu) + |u|2 + Vu2
t t
= (uu) 2 (u) ,

where " 2 #
1 u 2 2
(u) := + |u| + Vu
2 t
and denotes the gradient in the spatial coordinates. Note that
(u) assumes only positive values if V is a positive function.
Single Second-Order PDE

Example (cont.)
In addition,

2u
   
u u u
jk u, = u
t xk t t xk

Hence it follows that


 
u u
[ (uu) 2 (u)] = u u .
t t t
Single Second-Order PDE

Example (cont.)
The last implies that
 
u
(2 (u)) = 2 u
t t

and, finally, the conservation law


 
(u) u
= u . (7.28)
t t
Single Second-Order PDE

Example (cont.)
In physical applications, (u) is called the energy density and

u
u
t
the energy flux density associated with u.
Single Second-Order PDE
Example (cont.)
We note that (7.28) is true also for C 2 -solutions of (7.27) since
for such a u
 
(u) u
u
t t
"  #
2  
1 u 2 2 u
= + |u| + Vu u
t 2 t t
u 2 u
 
u u
= 2
+ (u) u + Vu u (u) 4u
t t t t t t
u 2 u
 
= 4u + Vu = 0 .
t t 2
Single Second-Order PDE

Problem 7.12
Let V C(, R) be such that

V
=0.
x1

Find a conservation law for all C 2 -solutions of (7.27).


Single Second-Order PDE

The following example shows that the above procedure for the
derivation of a conservation law can be generalized to nonlinear
equations.
Single Second-Order PDE

Example 29
(Energy conservation for a class of nonlinear Klein-Gordon
equations) We consider the case of a nonlinear Klein-Gordon
equation
2u
4u + m2 u + (V 0 u) = 0 , (7.29)
t 2
where m [ 0, ) denotes the mass of the field and the term
V 0 u describes a self-interaction of the field. Here
V C 4 (R, R).
Single Second-Order PDE

Example (cont.)
If u is a C 3 -solution of (7.29), then
 2 
u 2 0
0= 4u + m u + (V u)
t 2 t
2
ut
= 2 4ut + m2 ut + (V 00 u)ut ,
t
where a subscript t is used as a shorthand notation for a partial
time derivative.

Note that in general ut is no solution of (7.29). Still,


Single Second-Order PDE
Example (cont.)

2u
 
2 0
0 = ut 4u + m u + (V u)
t 2
 2 
u 2 0
u 4u + m u + (V u)
t 2
 2   2 t 
u 0 ut 00
= ut 4u + (V u) u 4ut + (V u)ut
t 2 t 2
= ut utt uuttt ut 4u + u4ut + [ 2(V u) u(V 0 u) ]t
= (ut2 uutt )t (ut u uut ) + [ 2(V u) u(V 0 u) ]t
= ut2 u4u + m2 u2 + 2(V u) t (ut u uut )
 
Single Second-Order PDE

Example (cont.)

= ut2 + m2 u2 + 2(V u) t ut 4u u4ut (ut u uut )


 

= ut2 + m2 u2 + 2(V u) t + |u|2 t (ut u + uut )


   

(ut u uut )
= ut + |u|2 + m2 u2 + 2(V u) t 2(ut u) .
 2 

As a consequence, we arrive at the conservation law,

(u)
= (ut u) (7.30)
t
Single Second-Order PDE

Example (cont.)
where
" 2 #
1 u
(u) := + |u|2 + m2 u2 + 2(V u)
2 t

and denotes the gradient in the spatial coordinates. In


physical applications, (u) is called the energy density and

u
u
t
the energy flux density associated with u.
Single Second-Order PDE
Example (cont.)
Note that (u) is not always positive-valued. Also, we note that
(7.30) is true for V C 3 (R, R) and a C 2 -solution u of (7.29)
since
 
(u) u
u
t t
"  #
2  
1 u 2 2 2 u
= + |u| + m u + 2(V u) u
t 2 t t
u 2 u
 
u 2 u u u
= + (u) + m u (u) 4u
t t 2 t t t t
u 0
+ (V u)
t
u 2 u
 
2 0
= 4u + m u + (V u) = 0 .
t t 2
Single Second-Order PDE

Figure 31: The domain of integration in the proof of Theorem 7.13 is


the region of the backward characteristic cone with apex x between
the parallel planes y0 = 0 and y0 = T .
Single Second-Order PDE

Figure 32: Sketch of the domain of integration for the case n = 2,


x1 = x2 = 0 in the proof of Theorem 7.13.
Single Second-Order PDE
Lemma 7.13
(Energy inequality) Let n {1, 2, 3}, x = (x0 , . . . , xn ) Rn+1
be such that x0 > 0. Further, let T [0, x0 ). We define the solid
backward characteristic cone SC x with apex x by
" n #1/2
X
SCx := y Rn+1
: y0 6 x 0 (x j yj )2
.

j=1

Finally, let be an open subset of Rn+1 such that

SC n
x ( [0, T ] R )

and u C 2 (, R) be a solution to (7.27), where V is assumed to


be everywhere positive on Cx ( [0, T ] Rn ).
Single Second-Order PDE

Lemma (cont.)
Then
Z Z
(u)(T , ) dy1 . . . dyn 6 (u)(0, ) dy1 . . . dyn .
Bxn (x1 ,...,xn ) Bxn0 (x1 ,...,xn )
0 T
(7.31)
Single Second-Order PDE
Proof.
I It follows from (7.28) by Gauss Theorem that
Z   
(u) u
0= u (y) dy
SCx ( [0,T ]Rn ) t t
Z  
u
=   (u), u dS
SCx ( [0,T ]Rn ) t
Z
= (u)(T , ) dy1 . . . dyn
Bxn (x1 ,...,xn )
Z0 T
(u)(0, ) dy1 . . . dyn
Bxn0 (x1 ,...,xn )
Z  
u
+ (u), u dS ,
Cx ( [0,T ]Rn ) t
Single Second-Order PDE

Proof (cont.).
I where denotes the outer unit normal field on the
boundary surface

SC n
of SC n

x ( [0, x0 ] R ) x ( [0, x0 ] R ) .

I A parametrization of Cx is given by f : Rn Rn+1 defined


" n #1/2
X
f (y1 , . . . , yn ) := x0 (xj yj )2 , y1 , . . . , yn
j=1

for every y = (y1 , . . . , yn ) Rn .


Single Second-Order PDE

Proof (cont.).
I In particular,
 
1 x1 y1 xn yn
(f (y)) = 1, ,...,
2 |x y| |x y|

for every y = (y1 , . . . , yn ) Rn ,


I where x := (x1 , . . . , xn ).
Single Second-Order PDE
Proof (cont.).
I In addition, it follows for such y that

  
u
2 2 (u), u ()
t
 2
u
= () + |(u)()|2 + V () (u())2
t
2 u
() (x y) (u)()
|x y| t
 2
u
> () + |(u)()|2 + V () (u())2
t

u
2 () |(u)|() > 0 ,
t

where := f (y). Hence it follows (7.31).


Single Second-Order PDE

Theorem 7.14
(Local Uniqueness) Let x = (x0 , . . . , xn ) Rn+1 be such that
x0 > 0. Further, let be an open subset of Rn+1 such that

SC n
x ( [0, x0 ] R )

and u, v C 2 (, R) be solutions of (7.27) assuming the same


values on {0} Btn (x1 , . . . , xn ), where it is assumed that
V (y) > 0 for all y . Then u and v coincide on
SC n
x ( [0, x0 ] R ).

Proof.
The statement follows by a simple application of Lemma (7.13)
to u v.
Single Second-Order PDE

Figure 33: Graph of u from Example 30. Here C = 1, T = L = 1 and


m = n = 3.
Single Second-Order PDE

x
O L

Figure 34: Visualization of in Example 30.


Single Second-Order PDE

Example 30
(Ill-posedness of the Dirichlet problem for the wave
equation) For this, let T > 0, L > 0 be such that T /L = m/n,
where m, n are elements of N . In addition, define
:= (0, T ) (0, L). Then for every C R, u : R defined
by
u(t, x) := C sin(mt/T ) sin(nx/L)
for every (t, x) is an element of C 2 (, R) satisfying

2u 2u
2 =0
t 2 x
Single Second-Order PDE

Example (cont.)
whose corresponding extension u to an element of C(, R)
assumes the boundary values

u(0, x) = u(T , x) = 0 , u(t, 0) = u(t, L) = 0 (7.32)

for every x [0, L] and t [0, T ]. As a consequence, the


boundary value problem corresponding to (7.32) for the wave
equation has infinitely many solutions and is not well-posed.
Problem 7.15
Problems 4.1, 4.2 of Chapter VIII, Section 4 in [23].
Single Second-Order PDE

I In the following, we introduce the method of separation of


variables in the special case of Laplaces equation in
Cartesian coordinates.
I The application of that method is not restricted to such
elliptic equations,
I but can be applied to linear PDE of all types, including
non-constant coefficients, and also to non-linear equations.
Single Second-Order PDE
I If applicable, the method reduces the process of finding
special solutions of the equation in form of a product to the
solution of a system of ordinary differential equations.
I Typically, in the case of non-linear equations, the method
is suitable only for the finding of particular solutions of the
equation,
I whereas in the case of linear equations a large family of
solutions is found that can be superposed to represent
solutions that do not separate.
I The proper tools for the treatment of such infinite sums
come from functional analysis, in particular, from the
theory of function spaces and operator theory, that are
subject of more advanced texts.
Single Second-Order PDE

I In order to find special solutions of

4u = 0

on some interval I1 In of Rn ,
I where n N \ {0, 1} and I1 , . . . , In are non-void open
subintervals of R,
I we make the product ansatz

u(x) = u1 (x1 ) un (xn ) (7.33)

for x = (x1 , . . . , xn ) I1 In with unknown twice


differentiable functions u1 : I1 R, . . . , un : In R.
Single Second-Order PDE

I Then

(4u)(x) = u100 (x1 )u2 (x2 ) . . . un (xn ) + + u1 (x1 )u2 (x2 ) . . . un00 (xn )
(7.34)

I Assuming for the moment that u has no zeros and at the


same time that u1 , . . . , un are three times differentiable,
I the latter is equivalent to
n
(4u)(x) X ui00 (xi )
= =0
u(x) i=1
ui (xi )

for every x = (x1 , . . . , xn ) I1 In .


Single Second-Order PDE

I This implies that


   00  0
4u ui
(x) = (xj ) = 0
xj u ui

for every x = (x1 , . . . , xn ) I1 In


I and hence the constancy of uj00 /uj for every j {1, . . . , n}.
I Hence there are separation constants C1 , . . . , Cn R such
that
u100 = C1 u1 , . . . , un00 = Cn un . (7.35)
Single Second-Order PDE

I Indeed, for all C1 , . . . , Cn R satisfying

C1 + + Cn = 0

I and any twice differentiable solutions


u1 : I1 R, . . . , un : In R satisfying (7.35),
I it follows by (7.34) that the corresponding u defined by
(7.33) satisfies Laplaces equation.
I The solutions of each of the individual equations of (7.35)
are either first order polynomials, products of an
exponential function, or a linear combination of sin and
cos functions.
Single Second-Order PDE
u

x
1

Figure 35: Graphs of u1 (x, 0.5), u3 (x, 0.5), u7 (x, 0.5) and u15 (x, 0.5)
from Example 31.
Single Second-Order PDE
y

x
-2 O 2

Figure 36: Sketch of the domain in Example (31).


Single Second-Order PDE

Example 31
(Hadamards example of an ill-posed initial value problem
for Laplaces equation) Let := (/2, /2) (0, ) and
n N be odd. Then an element un of C 2 (, R) satisfying

2 un 2 un
+ =0
x 2 y2
Single Second-Order PDE

Single Second-Order PDE


such that the corresponding continuous extensions

u
dn
un , C(, R)
y

of un and un /y assume the initial values

u
dn
un (x, 0) = 0 , (x, 0) = e n cos(nx)
y

for every x [/2, /2]


Single Second-Order PDE

Example (cont.)
and the boundary values

un (/2, y) = un (/2, y) = 0 (7.36)

for every y [0, ) is given by


1 n
un (x, y) := e cos(nx) sinh(ny)
n
for every (x, y) .
Single Second-Order PDE

Example (cont.)
In particular,
u
n
d
(x, 0) 6 e n

y


for every x [/2, /2] and hence

u
n
d
lim max (x, 0) = 0 .

n x[/2,/2] y
Single Second-Order PDE
Example (cont.)
Therefore, it might be expected that the sequence u1 , u3 , . . .
approaches in some sense the zero function
u := ( R, x 7 0) on which satisfies Laplaces equation
and is such that the corresponding continuous extensions

u
c
u, C(, R)
y

of u and u/y assume the initial values

u
c
u(x, 0) = 0 , (x, 0) = 0
y

for every x [/2, /2] and the boundary values (7.36).


Single Second-Order PDE
Example (cont.)
On the other hand, it follows for every y > 0 that

1 ne n
lim = lim =0.
n un (0, y) n sinh(ny)

Also
Z /2 Z /2
2 1 2n 2
u (x, y) dx = 2 e sinh (ny) cos2 (nx) dx
/2 n /2

Z /2
1
= 2 e2 n sinh2 (ny) [1 + cos(2nx)] dx
2n /2

= 2 e2 n sinh2 (ny)
2n
Single Second-Order PDE

Example (cont.)
and therefore
1
lim R =0.
n /2 u2 (x, y) dx
/2 n
Single Second-Order PDE

Example (cont.)
Hence small perturbations of the initial data of the zero solution
of Laplaces equation subject to the boundary conditions (7.36)
do not necessarily lead to small perturbations in the solution. As
a consequence, the corresponding initial boundary value
problem for Laplaces equation appears not to be well-posed.
Single Second-Order PDE

In general initial boundary value problems for elliptic PDE are


not well-posed. Usually, only boundary value problems
specifying data on the whole boundary of the corresponding
domain lead to well-posed problems.
Single Second-Order PDE

Figure 37: Plates of the capacitor from Prob 7.16.


Single Second-Order PDE

Problem 7.16
Let a, b > 0 such that a < b and let V1 , V2 R. Find a
harmonic function u C 2 (, R) on

:= (a, b) R2

that can be extended to a continuous function u on assuming


the boundary values
(
V1 for all x {a} R2
u(x, y) = .
V2 for all x {b} R2

The solution describes the potential field in an infinitely


extended parallel plate capacitor.
Single Second-Order PDE
y

W
W

x
O
Figure 38: 0 is the image of under a Euclidean transformation, a
rigid rotation around the origin about the angle /4.The black points
are the centers of the squares , 0 . Compare Lemma 7.17.
Single Second-Order PDE
Lemma 7.17
(Euclidean transformations leave the form of the Laplace
operator invariant) For this, let n N and Rn be
non-empty and open. In addition, let 0 Rn be a non-empty
open subset such that
g( 0 ) = .
Here g C (Rn , Rn ) is defined by

g(x) := Ax + b

for all x Rn , where A M(n n, R) is orthogonal, i.e.,


invertible and such that

At = A1 ,

and b Rn .
Single Second-Order PDE

Lemma (cont.)
Here At denotes the transposed matrix corresponding to A.
Finally, let u C 2 (, R) and u C 2 (0 , R) be defined by

u(x) := (u g)(x) = u(Ax + b)

for all x 0 . Then

(4u)(g(x)) = (4u)(x)

for all x 0 .
Single Second-Order PDE

Proof.
It follows by the chain rule that
n
u X u
(x) = Aki (g(x)) ,
xi k=1
x k
n
2 u X 2u
(x) = Aki Alj (g(x))
xi xj k,l=1
xk xl

for every x 0 and i, j {1, . . . , n}.


Single Second-Order PDE

Proof (cont.).
Hence
n X
n
X 2u
(4u)(x) = Aki Ali (g(x))
i=1 k,l=1
xk xl
n n
!
X X 2u
= Aki Ali (g(x)) = (4u)(g(x))
k,l=1 i=1
xk xl

for every x 0 .
Single Second-Order PDE

Remark 7.17.1
Note in particular in Lemma 7.17 that u is harmonic if and only
if u is harmonic.
Problem 7.18
Let n N . Prove that a real quadratic n n-matrix, n N , is
orthogonal if and only if the associated row vectors form a
orthogonal basis in Rn equipped with the Euclidean scalar
product.
Single Second-Order PDE

In the following, an orientation preserving C 1 -map between


non-void open subsets of Rn will be called conformal if it
preserves the oriented angles between tangent vectors of all
differentiable paths through points of its domain.
Single Second-Order PDE

W
W

Hg2 LH0L
gH2 HI2 LL
1 HI1 L gH1 HI1 LL
2 HI2 L

, Hg1 LH0L
, 1 H0L gHp L
2 H0L
p

Figure 39: Diagram on the properties of a conformal map g. Compare


Def. 7.22.
Single Second-Order PDE

Definition 7.19
Let n N , , 0 Rn be non-empty open subsets and
g C 1 ( 0 , ). Then g is called conformal if

det g 0 (p) > 0

for all p 0 and if there is a scale transformation


: 0 (0, ) such that for any p 0 and any two
differentiable paths 1 , 2 through p, i.e., from open subintervals
I1 and I2 , respectively, of R around 0 into 0 such that
1 (0) = 2 (0) = p,

(g 1 ) 0 (0) (g 2 ) 0 (0) = (p) 10 (0) 20 (0) .


Single Second-Order PDE
y y

6 x
0.1 0.3 0.4

5
-0.1

4
-0.2

3
-0.3
2

-0.4
1

-0.5
0 x
0 1 2 3 4 5 6

Figure 40: Parallels to the axes of a Cartesian coordinate system of


distances 1, . . . , 6 and their images under the transformation
g := (R2 \{0, 0} R2 \{0, 0}, (x, y) 7 (x/(x 2 +y2 ), y/(x 2 +y2 ))).
Since g is conformal, the images intersect orthogonally.
Single Second-Order PDE

Theorem 7.20
Let , 0 R2 be non-empty open subsets and g C 1 ( 0 , )
without critical points. Then g is conformal if and only if its
component functions g1 , g2 satisfy the Cauchy-Riemann
equations
g1 g2 g1 g2
= , = . (7.37)
x y y x
Single Second-Order PDE
Proof.
I For this, let p 0 and 1 : I1 0 , 2 : I2 0 be
differentiable paths from open subintervals I1 , I2 of R
around 0 such that 1 (0) = 2 (0) = p.
I In addition, let : 0 (0, ) and vi := i0 (0) for
i {1, 2}.
I It follows by the chain rule that

(g i ) 0 (0) = ( vi (g1 )(p), vi (g2 )(p) )


 
g1 g1 g2 g2
= (p) vi1 + (p) vi2 , (p) vi1 + (p) vi2
x y x y

for i {1, 2}.


Single Second-Order PDE
Proof (cont.).
I Hence

(g 1 ) 0 (0) (g 2 ) 0 (0)
  
g1 g1 g1 g1
= (p) v11 + (p) v12 (p) v21 + (p) v22
x y x y
  
g2 g2 g2 g2
+ (p) v11 + (p) v12 (p) v21 + (p) v22
x y x y
" 2  2 # " 2  2 #
g1 g2 g1 g2
= + (p) v11 v21 + +
x x y y
 
g1 g1 g2 g2
+ + (p) (v11 v22 + v12 v21 ) (7.38)
x y x y
Single Second-Order PDE
Proof (cont.).
I which gives
" 2  2 #
g1 g1
(g1 ) 0 (0)(g2 ) 0 (0) = + (p) 10 (0)20 (0)
x y

if g1 , g2 satisfy (7.37).
I In this case, it follows in addition that
 
0 g1 g2 g2 g1
det g (p) = (p)
x y x y
" 2  2 #
g1 g1
= + (p) > 0
x y
Single Second-Order PDE
Proof (cont.).
I and hence that g is conformal.
I Further, (7.38) leads to

(g 1 ) 0 (0) (g 2 ) 0 (0) (p) 10 (0) 20 (0)


" 2  2 #
g1 g2
= + (p) v11 v21
x x
" 2  2 #
g1 g2
+ + (p) v12 v22
y y
 
g1 g1 g2 g2
+ + (p) (v11 v22 + v12 v21 ) .
x y x y
Single Second-Order PDE

Proof (cont.).
I If g is conformal and is the corresponding scale
transformation,
I then the case that v1 = e1 and v2 = e2 leads to the equation
 
g1 g1 g2 g2
+ (p) = 0 (7.39)
x y x y
Single Second-Order PDE
Proof (cont.).
I and the cases that v1 = v2 = e1 , v1 = v2 = e2 to the
equations
" 2  2 #
g1 g2
(p) = + (p)
x x
" 2  2 #
g1 g2
= + (p) . (7.40)
y y

I From (7.39) follows the existence of R such that


g1 g2 g2 g1
(p) = (p) , (p) = (p)
y x y x
I and from (7.40) that {1, 1}.
Single Second-Order PDE

Proof (cont.).
I Further,
 
0 g1 g2 g2 g1
det g (p) = (p)
x y x y
" 2  2 #
g1 g2
= + (p) > 0
x x

I implies that = 1 and hence that g1 , g2 satisfy (7.37).


Single Second-Order PDE

Remark 7.20.1
Note that, as a consequence of Theorem 7.20, the component
functions g1 and g2 of every conformal g C 2 ( 0 , ) are both
harmonic , i.e., satisfy

2 g1 2 g1 2 g2 2 g2
+ = + =0,
x 2 y2 x 2 y2

where , 0 R2 are non-empty and open.


Single Second-Order PDE

Remark 7.20.2
The main source for conformal functions in two real variables
are holomorphic functions in one complex variable. The real
part u and imaginary part v of every holomorphic function
defined on a non-trivial open subset of R2 satisfy the
Cauchy-Riemann equations.
Single Second-Order PDE

Problem 7.21
Let f : 00 0 , g : 0 , h : 0 , where
, 0 , 00 R2 are non-empty and open, satisfy the
Cauchy-Riemann equations. Show that g, g + h, g h and g f
satisfy the Cauchy-Riemann equations, too, where R and
g h : 0 is defined by

(g h)(x, y) := (g1 h1 g2 h2 , g1 h2 + g2 h1 )(x, y)

for all (x, y) 0 .


Single Second-Order PDE

Problem 7.22
Prove that the map g : R2 \ {(1, 0)} R2 \ {(0, 1)} defined
by
1 x 2 y2
 
2y
g(x, y) := ,
(1 x)2 + y2 (1 x)2 + y2
for all (x, y) U1 (0) is bijective and conformal. In particular,
calculate g1 , g(U1 (0)), g(S 1 \ {(1, 0)}) and g(R2 \ B1 (0)).
Single Second-Order PDE

Lemma 7.23
(Transformation of the Laplace operator in two dimensions
under conformal maps) For this, let , 0 R2 be non-empty
and open and g C 2 ( 0 , ) be conformal. Finally, let
u C 2 (, R). Then
" 2  2 #1
g1 g1
(4u)(g(x)) = (x) + (x) (4u)(x)
x1 x2

for all x 0 , where u C 2 ( 0 , R) is defined by u := u g.


Single Second-Order PDE
Proof.
It follows by the chain rule that
2
u X gl u
(x) = (x) (g(x)) ,
xj l=1
xj xl
2
2 u X 2 gl u
(x) = (x) (g(x))
xi xj l=1
xi xj xl
2
X gl gk 2u
+ (x) (x) (g(x))
k,l=1
xj xi xk xl

2 2
X gk gl 2u X 2 gk u
= (x) (x) (g(x)) + (x) (g(x))
k,l=1
xi xj xk xl k=1
xi xj xk

for every x 0 and i, j {1, 2}.


Single Second-Order PDE
Proof (cont.).
Hence, it follows that
2 2 X 2
X 2 u X gk gl 2u
(4u)(x) = = (x) (x) (g(x))
i=1
xi2 i=1 k,l=1
xi xi xk xl
2 X 2 2
" 2 #
X 2 gk u X X gk gl 2u
+ 2
(x) (g(x)) = (x) (x) (g(x))
x i xk xi xi xk xl
i=1 k=1 k,l=1 i=1
2
" 2 # 2
" 2  #
X X 2 gk u X X gk 2 2 u
+ 2
(x) (g(x)) = (x) 2
(g(x))
x i xk xi x k
k=1 i=1 k=1 i=1
" 2  2 #
g1 g1
= (x) + (x) (4u)(g(x))
x1 x2

for all x 0 , where Remark 7.20.1, (7.39) and (7.40) have


been used.
Single Second-Order PDE

Problem 7.24
Calculate  2  2
g1 g1
+ ,
x1 x2
where g1 is the first component of the map g from Prob 7.22.
Single Second-Order PDE
y

gHPL
R
R 2r

P
r
-R R x

-R

Figure 41: The inversion g with respect to SR1 maps the point P onto
the point g(P) on the half-line through P originating from the origin.
In this, the distance of g(P) from the origin is given by R2 /r, where r
is the distance of P from the origin. All points on SR1 are fixed points
under g. Compare Lemma 7.25.
Single Second-Order PDE

Lemma 7.25
(Transformation of the Laplace operator under inversion
with respect to n-spheres) For this, let n N , R > 0 and
Rn \ {0} be non-empty and open. In addition, let
0 Rn \ {0} be a non-empty open subset such that

g( 0 ) = ,

where g C (Rn \ {0}, Rn \ {0}) is defined by

R2
g(x) := x (7.41)
|x|2

for all x Rn \ {0}.


Single Second-Order PDE
Lemma (cont.)
Note that g is an involution38 and that its fixed point set is given
by SRn . Finally, let u C 2 (, R) and u C 2 (0 , R) be defined
by
" 2n #
R2
 2 
2n R
u(x) := u (g(x)) = |x| u x
|| |x|2

for all x 0 . Then


|x|n+2
(4u)(g(x)) = (4u)(x)
R4
for all x 0 .

38
i.e, g(g(x)) = x for all x Rn \ {0},
Single Second-Order PDE

Proof.
See the proof of Lemma 8.1 in the Appendix.
Remark 7.25.1
The inversion with respect to n-spheres allows the reduction of
some boundary value problems for Poissons equation on
infinitely extended domains to such problems on finitely
extended domains. Note that this inversion is not conformal.
Single Second-Order PDE

y y

1
2

u =1

u HWL

u =0
1
x
x 1
1 2

Figure 42: Diagram for Prob. 7.26.


Single Second-Order PDE

Problem 7.26
Problem 12.8 of Chapter VII, Section 12 in [23].
Single Second-Order PDE

Combined with the method of separation of variables, the


following transformations of the Laplace operator into special
coordinates, such as polar, cylindrical and spherical coordinates,
simplify the solution of boundary value problems for Poissons
equation displaying a special symmetry.
Single Second-Order PDE
Lemma 7.27
(Transformation of the Laplace operator in two dimensions
into polar coordinates) For this, let R2 \ {(0, 0)} be
non-empty and open. In addition, let p R2 be a non-empty
open subset such
g(p ) = ,
where g C (R2 , R2 ) is defined by

g(r, ) := (r cos(), r sin())

for all (r, ) R2 . Finally, let u C 2 (, R). Then

2 u 1 u 1 2 u
(4u)(g(r, )) = (r, ) + (r, ) + (r, )
r 2 r r r 2 2

for all (r, ) p ,


Single Second-Order PDE

Lemma (cont.)
where u C 2 (p , R) is defined by

u(r, ) := (u g)(r, ) = u(r cos(), r sin())

for all (r, ) p .


Proof.
See the proof of Lemma 8.2 in the Appendix.
Single Second-Order PDE
j y


V
2

V W
Wp

0 r x
0 0

Figure 43: Sketch of the regions p and from Problem 32. In


addition, the values of u, u are indicated on the boundaries.
Single Second-Order PDE
y

1
2

0 1
x
0 2
1

Figure 44: Contour plot of u from Problem 32 in the case V = 1.


Darker areas indicate higher function values.
Single Second-Order PDE

Example 32
Let V R. Find a solution u C 2 (, R) of the Laplace
equation on := (0, )2 that can be extended to a continuous
function u on \ {(0, 0)} assuming the boundary values

u(x, 0) = 0 , u(0, y) = V

for all x, y (0, ).


Single Second-Order PDE

Example (cont.)
Solution: The main idea is to adapt the coordinates in such a
way that is transformed into a strip in R2 . Polar coordinates
are suitable for this. As above, we define g C (R2 , R2 ) by

g(r, ) := (r cos(), r sin())

for all (r, ) R2 . Then

g(p ) = ,

where
p := (0, ) (0, /2) .
Single Second-Order PDE
Example (cont.)
Then u := u g|p has to satisfy

2 u 1 u 1 2 u
[ (4u)(g(r, )) = ] (r, )+ (r, )+ (r, ) = 0
r 2 r r r 2 2

for all (r, ) p and be extendable to a continuous function b


u
on (0, ) [0, /2] satisfying

u(r, 0) = 0 , b
b u(r, /2) = V

for all r > 0. Such u is given by


2V
u(r, ) :=

for all (r, ) p .
Single Second-Order PDE

Example (cont.)
Therefore u C (, R) defined by
!
2V x
u(x, y) := arccos p
x 2 + y2

for every (x, y) has the required properties.


Single Second-Order PDE

Problem 7.28
Problem 12.9 of Chapter VII, Section 12 in [23].
Single Second-Order PDE

Lemma 7.29
(Transformation of the Laplace operator in three
dimensions into cylindrical coordinates) For this, let
R3 \ ({0} {0} R) be non-empty and open. In
addition, let cyl R3 be a non-empty open subset such

g(cyl ) = ,

where g C (R3 , R3 ) is defined by

g(r, , z) := (r cos(), r sin(), z)

for all (r, , z) R3 . Finally, let u C 2 (, R).


Single Second-Order PDE
Lemma (cont.)
Then
2 u 1 u 1 2 u
(4u)(g(r, , z)) = (r, , z) + (r, , z) + (r, , z)
r 2 r r r 2 2
2 u
+ 2 (r, , z)
z
for all (r, , z) cyl , where u C 2 (cyl , R) is defined by

u(r, , z) := (u g)(r, , z) = u(r cos(), r sin(), z)

for all (r, , z) cyl .


Proof.
See the proof of Lemma 8.3 in the Appendix.
Single Second-Order PDE

Figure 45: is the region between 2 concentric cylinders. Compare


Example 32. Prob. 7.30.
Single Second-Order PDE

Problem 7.30
Let a, b > 0 be such that a < b and let u1 , u2 R. Find a
harmonic function u C 2 (, R) on

:= (Ub (0) R) \ (Ba (0) R)

that can be extended to a continuous function u on assuming


the boundary values
(
u1 for all x Sa1 R
u(x) = .
u2 for all x Sb1 R

The solution describes the potential field in an infinitely


extended cylindrical capacitor.
Single Second-Order PDE

Lemma 7.31
(Transformation of the Laplace operator in three
dimensions into spherical coordinates) For this, let
R3 \ ({0} {0} R) be non-empty and open. In
addition, let sph R3 be a non-empty open subset such

g(sph ) = ,

where g C (R3 , R3 ) is defined by

g(r, , ) := (r sin() cos(), r sin() sin(), r cos())

for all (r, , z) R3 . Finally, let u C 2 (, R).


Single Second-Order PDE
Lemma (cont.)
Then
2 u 2 u
(4u)(g(r, , z)) = 2 (r, , z) + (r, , z)
 2 r r r
1 u 2 2 u
+ 2 2 (r, , z) + sin () (r, , z)
r sin () 2 2

u
+ sin() cos() (r, , z)

for all (r, , z) sph , where u C 2 (sph , R) is defined by

u(r, , ) := (u g)(r, , )
= u(r sin() cos(), r sin() sin(), r cos())

for all (r, , ) sph .


Single Second-Order PDE

Proof.
See the proof of Lemma 8.4 in the Appendix.
Single Second-Order PDE

Figure 46: is the region between 2 concentric spheres. Compare


Prob. 7.32.
Single Second-Order PDE
Problem 7.32
Let a, b > 0 be such that a < b, A, B R, u1 , u2 R and
R3 be defined by := Ub (0) \ Ba (0). Find u C 2 (, R)
such that
B
(4u)(x) = A +
|x|
for every x that can be extended to a continuous function u
on assuming the boundary values
(
u1 for all x Sa2
u(x) = .
u2 for all x Sb2

Problem 7.33
Prove Lemmata 7.27, 7.29 and 7.31.
Single Second-Order PDE
y

x
HIL

Figure 47: Sketch of structures in the motivation of maximum


principles.
Single Second-Order PDE

The classical tool for proving uniqueness and continuous


dependence on the boundary data of the solutions of boundary
value problems for elliptic PDE are maximum principles. In the
following, we shall only discuss strong maximum principles.
Single Second-Order PDE

For motivation, let be a non-trivial open subset of Rn , where


n N , and u C 2 (, R) such that

4u > 0 . (7.42)

As a consequence, u has no local maximum.


Single Second-Order PDE
This can be seen as follows.
I Assume that u assumes a local maximum in p .

I Further, let v Rn an define the C -path : I by

( ) := p + v

for every I,
I where I is some sufficiently small open interval I around 0.
I Then it follows by the chain rule that
n n
0
X u 0
X u
(u ) ( ) = (( )) i ( ) = vi (( )) ,
i=1
xi i=1
xi
n
X 2u
(u ) 00 ( ) = (( )) vi vj
i,j=1
xi xj

for every I.
Single Second-Order PDE
I In particular, (u ) 0 (0) = 0 since p is a critical point of u.
I Since u assumes a local maximum value in p, u
assumes a local maximum value in 0
I and hence39
n
00
X 2u
(u ) (0) = (( )) vi vj 6 0 .
i,j=1
xi xj

I This implies that


2u
(p) 6 0
xi2
for all i {1, . . . , n}.
I The latter leads to the contradiction that (4u)(p) 6 0.
I Therefore, u has no local maximum.
39
See, e.g., Theorem 1.83 in [2] and the corresponding proof.
Single Second-Order PDE

In the following, we generalize the previous reasoning to a


much wider class of elliptic differential operators and also to
include the case of equality in (7.42) which is of main interest.
Single Second-Order PDE
I For this, we consider the linear second-order partial
differential operator A defined by
n n
X 2u X u
(Au)(x) := aij (x) (x) + bi (x) (x) + c(x) u(x)
i,j=1
xi xj i=1
xi

for every x .
I We assume that aij : R, i, j {1, . . . , n}, satisfy

aij = aji

for all i, j {1, . . . , n} and x .


I Further, we assume that bi : R for i {1, . . . , n},
c:R
I and that u C 2 (, R).
Theorem 7.34
In addition, let c be such that c(x) 6 0 for all x , let
n
X
aij (x) vi vj > 0 (7.43)
i,j=1

for 1309 all v Rn \ {0} and x . Finally, let u C 2 (, R)


be such that
(Au)(x) > 0
for all x . Then there is no positive local maximum for u.
Single Second-Order PDE
Proof.
I The proof is indirect.
I We assume that there is p such that u(p) > 0 and
u(p) > u(q) for all q in some open neighborhood U
of p.
I Then p is a critical point of u.
I Further,
n
X 2u
aij (p) (p) = (Au)(p)c(p) u(p) > (Au)(p) > 0 .
i,j=1
x i xj
(7.44)
n
In addition, for all v R :
n
X 2u
(p) vi vj 6 0 .
i,j=1
xi xj
Single Second-Order PDE
Proof (cont.).
I We define the symmetric n n matrices a(p), H(p) by

2u
(a(p))ij := aij (p) , (H(p))ij := (p)
xi xj

for all i, j {1, . . . , n}.


I Then
n n
X 2u X 2u
aij (p) (p) = aij (p) (p) = Tr(a(p)H(p)) ,
i,j=1
xi xj i,j=1
xj xi

I where Tr := (M(n n, R) R, B B11 + + Bnn ) is


the trace function.
Single Second-Order PDE

Proof (cont.).
I According to linear algebra, there exists an orthogonal
matrix U such that

U a(p) t U 1 = diag(1 , . . . , n ) ,

I where 1 , . . . , n denote the eigenvalues of a(p)


I and diag(1 , . . . , n ) the diagonal matrix containing these
eigenvalues in the diagonal.
I As a consequence of (7.43), these eigenvalues are all > 0.
Single Second-Order PDE
Proof (cont.).
I For this, note that for every real symmetric n n-matrix B
satisfying
Xn
Bij vi vj > 0 (> 0)
i,j=1
n
for all v R \ {0} and every orthogonal real
n n-matrix C,
I it follows that
n
X n
X
t
(C B C)ij vi vj = Bij (tCv)i (tCv)j > 0 (> 0)
i,j=1 i,j=1

for all v Rn \ {0}.


Single Second-Order PDE

Proof (cont.).
I Finally,

Tr(a(p)H(p)) = Tr U 1 diag(1 , . . . , n )UH(p)




= Tr U 1 diag(1 , . . . , n )UH(p)U 1 U


= Tr diag(1 , . . . , n )UH(p)U 1


Xn
= i (UH(p)t U 1 )ii 6 0
i=1

I which is in contradiction to (7.44).


Single Second-Order PDE

Problem 7.35
Let n N and A, B, U real n n-matrices. In particular, let U
be invertible. Prove that

Tr(AB) = Tr(BA) , Tr(UAU 1 ) = Tr(A) .


Single Second-Order PDE
Theorem 7.36
(E. Hopf) In addition, let be pathwise connected40 and for
every x and every open ball Ux around x let there exist
x > 0 such that
Xn
aij (y) vi vj > x |v|2 (7.45)
i,j=1

for all v Rn and y Ux .41 Further, let aij , bj , i, j {1, . . . , n},


c be bounded on every bounded subset of and c(x) 6 0 for all
x . Finally, let u C 2 (, R) satisfy

(Au)(x) > 0

for all x and assume a positive maximum. Then u is a


constant function.
40
An open subset of Rn is pathwise connected if and only if it is connected.
On the other hand, if applicable, the direct proof of pathwise connectedness
Single Second-Order PDE
B

p
B

Figure 48: Sketch in the proof of Theorem 7.36. Here B := U and


B 0 := B 0 (p).
Single Second-Order PDE

Proof.
I The proof is indirect.
I We assume that u is non-constant.
I In the first step, we prove that there is an open ball U
around a q such that U ,

u(q) < (7.46)

for all q U
I and for which there is p U such that

u(p) = .
Single Second-Order PDE

Proof (cont.).
I For the construction of such U,
I let > 0 be the maximum value of u,
I p1 such that u(p1 ) = ,
I q1 such that u(q1 ) <
I and : [0, 1] be a continuous path such that
(0) = q1 and (1) = p1 .
I Further, let

t2 := min 1 () , p2 := (t2 ) .

I Then t2 > 0, u(p2 ) = and u(t) < for all t [0, t2 ).


Single Second-Order PDE

Proof (cont.).
I In addition, let d > 0 be the distance between Ran and
.
I Further, let q Ud/2 (p2 ) ([0, t2 )) and U the largest open
ball around q that is contained in u1 ((, )).
I The radius of this ball is less or equal to d/2 and hence
U .
I This U satisfies (7.46) and, in particular, there is p
such that u(p) = .
Single Second-Order PDE
Proof (cont.).
I Further, let r be a point in the open line segment between p
and q,
I
:= |r p| (> 0)
I and 0 (0, ) be such that

B 0 (p) .

I Then

0 < 0 6 |xp| = ||xp|| 6 |rp(xp)| = |rx|

for all x B 0 (p).


Single Second-Order PDE
Proof (cont.).
I We define
2 2 2 2
v(x) := ek|rx| ek = ek|rx| ek|rp|

for every x , where k [0, ).


I Then
( n
2
X
(Av)(x) = ek|rx| 4k 2 aij (x)(xi ri )(xj rj )
i,j=1
n
)
X 2
2k [aii (x) + bi (x)](xi ri ) + c(x) c(x) ek
i=1

for all x .
Single Second-Order PDE
Proof (cont.).
I Obviously, it follows that

(Av)(x) > 0

for all x B 0 (p) for sufficiently large k.


I Such k is assumed in the following.
I As a consequence for > 0

[A(u + v)](x) > 0

for all x B 0 (p).


I Also
(u + v)(p) = u(p) .
Single Second-Order PDE
Proof (cont.).
I Further, let x B 0 (p).
I Then there are two cases.
I In the first case
|x r| >
I and hence v(x) < 0
I as well as
(u + v)(x) < u(x) 6 u(p) .
I In the second case, |x r| 6 , x U
I and therefore

(u + v)(x) < u(p) + v(x) .


Single Second-Order PDE

Proof (cont.).
I Hence it follows for small enough that

(u + v)(x) < u(p)

for all x B 0 (p).


I Obviously, from this follows that for sufficiently small
> 0 the restriction of u + v to U 0 (p) has a positive
maximum in p.
I This is a contradiction to Theorem (7.34).
I Hence u is a constant function.
Single Second-Order PDE

Corollary 7.37
In addition, to the assumptions of Theorem 7.36, assume that
c = 0. If u assumes a maximum, then u is a constant function.
Proof.
Assume that u has a maximum in the point p U. Then
v := u u(p) C 2 (, R) has a positive maximum at p and
satisfies
(Av)(x) > 0
for all x . Hence it follows by Theorem 7.36 that v and
therefore also u are constant functions.
Single Second-Order PDE

Example 33
Define u : R2 R by

u(x, y) := 1 (x 2 + y2 )

for every (x, y) R2 . Then

2u 2u
2 =0.
x 2 y
Also, u is non-constant and assumes a positive maximum.
Hence the statement of Theorem 7.36 is in general false for
hyperbolic A.
Single Second-Order PDE

Example 34
Let n N , a > 0. Define u : Rn R by

u(x) := a

for every x Rn . Then


4u = 0 ,
and u assumes a positive maximum. Hence the reference to
constant functions in the statement of Theorem 7.36 is
important.
Single Second-Order PDE

Example 35
Define u : R2 R by

u(x, y) := (x 2 + y2 ) 4

for every (x, y) R2 . Then

(4u u)(x, y) = x 2 + y2 > 0 ,

and u is non-constant and assumes a strictly negative maximum.


This example illustrates the importance of the assumption of
positivity of the maximum in Theorem 7.36.
Single Second-Order PDE

Example 36
Define u : U1 (0) R by

u(x, y) := (x 2 + y2 ) + 5

for every (x, y) U1 (0). Then

(4u + u)(x, y) = 1 (x 2 + y2 ) > 0 ,

and u is non-constant and assumes a positive maximum. Hence


the statement of Theorem 7.36 would be false in general if the
assumption that c 6 0 would be omitted.
Single Second-Order PDE

Lemma 7.38
Let be a non-empty pathwise connected open subset of Rn
and u C 2 (, R) be non-constant and such that42

4u = 0 .

Then u does not assume a maximum or a minimum.


Proof.
It follows by Corollary 7.37 that neither u nor u assume a
maximum. As a consequence, u does not assume a maximum or
a minimum.

42
Functions satisfying the following equation are called harmonic.
Single Second-Order PDE

Problem 7.39
Problems 1.1 - 1.3 of Chapter VII, Section 1 in [23].
Single Second-Order PDE

In the following, we prove uniqueness of the classical solutions


of certain elliptic equations as consequences of Hopf strong
maximal principle. For this, we define the subspace C (Rn , R)
of C(Rn , R) to consist of those functions f that in addition
vanish at infinity, i.e, satisfy

lim |f (x )| = 0

for all sequences x0 , x1 , . . . in Rn \ {0} such that


1
lim =0.
|x |
Single Second-Order PDE

p
q
fHpL O fHqL

Figure 49: The points p, q on the unit circle are mapped to the points
f (p), f (q) under the stereographic projection. Compare the proof of
Lemma 7.40.
Single Second-Order PDE
Lemma 7.40
Every element of C (Rn , R) that assumes a positive value also
assumes a maximum.
Proof.
I For this, we define the stereographic projection
g C(S n+1 \ {N}, Rn ) from the north pole N := en+1 by
 
p1 pn
g(p) := ,...,
1 pn+1 1 pn+1
I In particular, g is bijective with the continuous inverse
g1 : Rn S n+1 \ {N} given by

2x1 2xn |x|2 1


g1 (x) = e 1 + . . . en + en+1 .
|x|2 + 1 |x|2 + 1 |x|2 + 1
Single Second-Order PDE
Proof (cont.).
I Then
h := f g C(S n+1 \ {N}, R) .
I We extent h to a function h : S n+1 R by defining
h(N) := 0 (and h(p) := h(p) for every p S n+1 \ {N}.
I Obviously, if q1 , q2 . . . is a sequence in
S n+1 \ ({N} {en+1 }) that converges to N,
I then
1
lim =0
n |g(qn )|

I Hence h is continuous.
I Since S n+1 is compact, h assumes a maximum and a
minimum value.
Single Second-Order PDE

Proof (cont.).
I Since f assumes a positive value,
I this maximum is assumed in S n+1 \ {N}.
I Therefore, f assumes a maximum value.
Single Second-Order PDE
y

x
-10 -2 2 10

-1

Figure 50: Graph of f := (R R, x 7 1/(1 + x 2 )).


f C (R, R), but does not assume a maximum. Compare
Example 37.
Single Second-Order PDE

Example 37
The function f : R R defined by
1
f (x) :=
1 + x2
for every x R is an element of C (R, R) that assumes no
maximum value. Hence the statement of Lemma 7.40 would be
false in general without the assumption that f assumes a positive
value.
Single Second-Order PDE

Theorem 7.41
Let u C 2 (Rn , R) C (Rn , R) such that

4u = 0 .

Then u(x) = 0 for all x Rn .


Single Second-Order PDE
Proof.
I The proof is indirect.
I We assume that u does not vanish everywhere.
I Since u C (Rn , R), either u or u assume a maximum
according to Lemma 7.40.
I Hence u assumes a maximum or a minimum.
I Also, since u is an element of C (Rn , R) that does not
vanish everywhere,
I it follows that u is not a constant function.
I Hence, according to Lemma 7.38, u does not assume a
maximum or a minimum.
Single Second-Order PDE

Example 38
Examples of non-trivial harmonic functions that are not
vanishing at infinity are easy to find. For instance, such function
is given by u : R2 R defined by u(x, y) := x 2 y2 .
Single Second-Order PDE

Corollary 7.42
(Uniqueness of the classical Dirichlet problem) Let
u, v C 2 (Rn , R) C (Rn , R) such that

4u = 4v .

Then u = v.
Proof.
By the assumptions, u v C 2 (Rn , R) C (Rn , R) is
harmonic. Hence it follows by Theorem 7.41 that u v
vanishes everywhere and hence that u = v.
Single Second-Order PDE

Theorem 7.43
(A priori estimates) In addition, let be pathwise connected
and bounded. Also, let a R and d > 0 such that
(a, a + d) Rn1 . Further, we assume that
aij , bi , c C(, R) for all i, j {1, . . . , n} and that c(x) 6 0 for
all x . In addition,43 let there be C(, (0, )) such that
n
X
aij (x) vi vj > (x)|v|2
i,j=1

for all v Rn and x .

43
In the following, there will be used the same symbols for functions
defined on and their continuous extensions to . It will be clear from the
context whether the function itself or its extension is referred to.
Single Second-Order PDE

Theorem (cont.)
Finally, let u C 2 (, R) C(, R) such that

Au = f ,

where f C(, R). Then

|f (x)|
max |u(x)| 6 3 max |u(x)| + 2C max ,
x x x (x)
where
|b(x)|
C := exp(( + 1)d) 1 , := max .
x (x)
Single Second-Order PDE
Proof.
I For this, we define the auxiliary function h C (, R)
by
h (x) := exp((x1 a))
for every x = (x1 , . . . , xn ) and := 1 + .
I Then for every x :
n n
X 2 h X h
[(A c) h )](x) = aij (x) (x) + bi (x) (x)
i,j=1
xi xj i=1
xi

2

= a11 (x) + b1 (x) h (x)
 
2 |b1 (x)|
> a11 (x) || (x) h (x)
(x)
> ||(|| ) (x) h (x) = (1 + ) (x) h (x) > (x) .
Single Second-Order PDE

Proof (cont.).
I We define v C (, R) by

v(x) := max |u| + f (ed h (x))


x

for every x ,
I where
|f (x)|
f := max .
x (x)
Single Second-Order PDE
Proof (cont.).
I Then v(x) > 0 for all x
I and
v(x) > u(x)
for all x and

[A(v u)](x) = [(A c)v](x) f (x) + c(x) v(x)


6 [(A c)v](x) f (x) = f [(A c)h ](x) f (x)
6 f (x) f (x) 6 (|f (x)| + f (x)) 6 0

for all x .
I Hence according to Theorem 7.36, the function v u is
either constant or does not assume a positive maximum
Single Second-Order PDE
Proof (cont.).
I and therefore

(vu)(x) 6 max |v(x)u(x)| 2 max |u(x)|+f (ed 1)


x x

I and

u(x) = v(x) + (u v)(x) 6 v(x) 6 3 max |u| + 2f (ed 1)


x

for every x .
I Further, since
A(u) = f ,
applying the previous result to u,
Single Second-Order PDE

Proof (cont.).
I it follows that

u(x) 6 3 max |u| + 2f (ed 1)


x

for every x and,


I finally, that

|u(x)| 6 3 max |u| + 2f (ed 1)


x

for all x .
Single Second-Order PDE
Corollary 7.44
(Uniqueness and continuous dependence on the boundary
data of the solutions of the classical Dirichlet problem) In
addition to the assumptions of Theorem 7.43, let
v C 2 (, R) C(, R) be such that

Av = f .

Then
max |u(x) v(x)| 6 3 max |u(x) v(x)| ,
x x

where u : R and v : R are continuous extensions of u


and v, respectively. In particular, it follows that u = v if

u(x) = v(x)

for every x .
Single Second-Order PDE

Proof.
The statements follows by a simple application of Theorem 7.43
to u v.
Single Second-Order PDE

In the following, we derive Greens representation formula for


elements of C 2 (, R) C 1 (, R), where Rn is a
non-empty bounded open subset for which Gauss theorem is
valid and n {2, 3}.
Single Second-Order PDE

Theorem 7.45
Let n {2, 3}, Rn be a non-empty bounded open subset
for which Gauss theorem is valid, u C 2 (, R) C 1 (, R)
and x . Then44
Z Z
u(x) = x 4u dy + [ u (x ) x (u) ] dSy ,

(7.47)

44
In the following, the same symbols will be used for functions defined on
open sets and their extensions to continuous functions defined on the
closures of those sets. It will be clear from the context whether the function
itself or its extension is referred to.
Single Second-Order PDE

Theorem (cont.)
where x C (Rn \ {x}, R) is defined by
(
|x y|2n / [n(2 n)v(B1 (0))] for n > 2
x (y) :=
ln(|x y|) / (2) for n = 2

for every y Rn \ {x},


Z
v(B1 (0)) := dy
B1 (0)

is the volume of the unit ball in Rn and : Rn is the outer


unit normal field on .
Single Second-Order PDE

Proof.
I For this, we define
(
r 2n / [n(2 n)v(B1 (0))] for n > 2
(r) := (7.48)
ln(r) / (2) for n = 2

for every r (0, ).


I Then x (y) = (|x y|) for every y Rn \ {x}.
Single Second-Order PDE
Proof (cont.).
I Further,
x 1
(y) = (yi xi ) |x y|n ,
yi nv(B1 (0))
2 x 1
ij |x y|n

(y) =
yi yj nv(B1 (0))
n(xi yi )(xj yj )|x y|(n+2)


for all y Rn \ {x}, i, j {1, . . . , n},


I where (
0 if i 6= j
ij := .
1 if i = j
Single Second-Order PDE
Proof (cont.).
I As a consequence,

(4x )(y)
n
1 X
|x y|n n(xi yi )2 |x y|(n+2) = 0
 
=
nv(B1 (0)) i=1

I and

x 1 1n
xi (y) 6 nv(B1 (0)) |x y| ,

2
x 1 n
xi xj 6 v(B1 (0)) |x y|
(y)

for all y Rn \ {x}, i, j {1, . . . , n}.


Single Second-Order PDE

Proof (cont.).
I It follows for every non-empty bounded open subset
0 Rn for which Gauss theorem is valid
I and v C 2 ( 0 , R) C 1 ( 0 , R) that
Z Z
(v4u u4v) dx = (vu uv) dx
0
Z 0

= [ v (u) u (v) ] dSy (7.49)


0
Single Second-Order PDE

Proof (cont.).
I In particular,
Z Z
x 4u dy = [ x (u) u (x ) ] dSy
\B (x)
Z
+ [ x (u) u (x ) ] dSy .
S (x)
Single Second-Order PDE
Proof (cont.).
I Since
Z ! Z

x (u) dSy 6 sup |u| x dSy


S (x) yB (x) S (x)
! Z
= sup |u| () dSy ,
yB (x) S (x)

I it follows that
Z

lim x (u) dSy = 0 . (7.50)

0 S (x)

Single Second-Order PDE

Proof (cont.).
I Also,
Z Z
0
u (x ) dSy = () u dSy
S (x) S (x)

implies that
Z
lim u (x ) dSy = u(x) . (7.51)
0 S (x)

I Finally, it follows the representation (7.47).


Single Second-Order PDE

Problem 7.46
Fill in the details that lead to (7.50), (7.51) and the final
conclusion in the previous proof.
Single Second-Order PDE

Corollary 7.47
In addition to the assumptions of Theorem 7.45, let h : 2 R
be such that hx := h(x, ) is harmonic as well as an element of
C 2 (, R) C 1 (, R) for every x . Further, we define
Gx := x + hx for all x . Then
Z Z
u(x) = Gx 4u dy + [ u (Gx ) Gx (u) ] dSy .

(7.52)
Single Second-Order PDE

Proof.
It follows by (7.49) that
Z Z
hx 4u dy = [ hx (u) u (hx ) ] dSy
0 0

and hence by (7.47) the representation (7.52).


Single Second-Order PDE

Remark 7.47.1
If, in addition to the assumptions of Theorem 7.47, Gx admits an
extension Gx C( \ {x}, R) that vanishes on for every
x , then we call G the (Dirichlet) Greens function of the
first kind for . In this case
Z Z
u(x) = Gx 4u dy + u (Gx ) dSy

for every x .
Single Second-Order PDE
In the following, we solve the Dirichlet problem for open balls
around the origin.
Theorem 7.48
(Poissons integral) Let n {2, 3} and R > 0. We define
D := {(x, y) R2n : x 6= y}, := UR (0) and
G C (2 \ D, R) by
s
 2
|x||y|
G(x, y) := x (y) + R2 2x y
R
s
 2
p  |x||y|
= |x|2 + |y2 | 2x y + R2 2x y
R

for all (x, y) 2 \ D, where C (Rn \ {0}) is defined by


(7.48).
Theorem (cont.)
Then,
(i) G is the (Dirichlet) Greens function of the first kind for .
In particular,

G(x, y) = G(y, x) , G(x, y) < 0

for all (x, y) 2 \ D.


Theorem (cont.)
(ii) In addition, let f : R be continuous. Then by
Poissons integral

R2 |x|2
Z
f (y)
u(x) := dSy
nRv(B1 (0)) |x y|n

for every x there is defined a harmonic element


u C 2 (, R) C(, R) such that

lim u(x) = f (x0 )


xx0

for all x0 .
Single Second-Order PDE
Proof.
(i):
I First, we note that
 2  2
|x||y| 2 |x||y|
+R 2xy = R +2 (|x||y|xy) > 0
R R

for all x , y
I and hence that h : 2 R defined by
s
 2
|x||y|
h(x, y) := + R2 2x y
R

for every (x, y) 2


Single Second-Order PDE

Proof (cont.).
I is well-defined as well as an element of
C (, R) C 1 (, R).
I As a consequence of
" 2 #
|x||y|
+ R2 2x y |x|2 + |y|2 2x y
 
R
 2
2 2 2 |x||y|
= R (|x| + |y| ) + >0
R

for all (x, y) 2 ,


Single Second-Order PDE
Proof (cont.).
I it follows that
G(x, y) < 0
for all (x, y) 2 \ D.
I Also, obviously,
G(x, y) = G(y, x)
for all (x, y) D
I and for every x , Gx := G(x, ) admits an extension to a
function Gx C(, R) such that

Gx (y) = 0

for all y .
Single Second-Order PDE

Proof (cont.).
I In the following, g denotes the map defined by (7.41)
describing inversion with respect to SRn .
I In the case n > 2, it follows for x Rn and y Rn \ {0, x}
that
Single Second-Order PDE

Proof (cont.).

 2n
R
x (y) x (g(y))
|g(y)|
 2n
R
= (|x y|) (|x g(y)|)
|g(y)|
"  2n 2n #
1 2n |y| R2
= |x y| x 2 y
n(2 n)v(B1 (0)) R |y|
" 2n #
1 |y| R
= |x y|2n x y
n(2 n)v(B1 (0)) R |y|
Single Second-Order PDE

Proof (cont.).


1 p 2n
= |x|2 + |y2 | 2x y
n(2 n)v(B1 (0))
s 2n
 2
|x||y|
+ R2 2x y
R
s
 2
p  |x||y|
= |x|2 + |y2 | 2x y + R2 2x y
R

= G(x, y)
Single Second-Order PDE
Proof (cont.).
I and hence that h(x, ) is harmonic for every x .
I In the case n = 2, it follows for x Rn and y Rn \ {0, x}
that
 
1 |y|
(|x y|) (|x g(y)|) ln
2 R
2
 
1 1 R 1 |y|
= ln |x y| ln x 2 y
ln
2 2 |y| 2 R

1 1 |y| R
= ln |x y| ln x y
2 2 R |y|
s
q   2
|x||y|
= |x|2 + |y2 | 2x y + R2 2x y
R

= G(x, y)
Single Second-Order PDE

Proof (cont.).
I and hence that h(x, ) is harmonic for every x .
(ii):
I According to Remark 7.47.1 it follows for every
v C 2 (, R) C 1 (, R) that
Z Z
v(x) = Gx 4v dy + v (Gx ) dSy (7.53)

for every x ,
I where : Rn is the outer unit normal field on .
Single Second-Order PDE

Proof (cont.).
I Further,
Gx 1
(y) = (yi xi ) |x y|n
yi nv(B1 (0))
 
1 1
+ 2 yi |x| + xi |x y|n
2
nv(B1 (0)) R
1
R2 |x|2 yi |x y|n

= 2
nR v(B1 (0))
Single Second-Order PDE

Proof (cont.).
I and hence
n
X yi Gx
(Gx )(y) = (y)
i=1
|y| yi

1
R2 |x|2 |x y|n

=
nRv(B1 (0))

for all x and y .


Single Second-Order PDE

Proof (cont.).
I In particular, the special case that v(x) = 1 for all x of
(7.53) leads to Z
K(x, y) dSy = 1

for every x ,
I where Poissons kernel K : R is defined by
1
R2 |x|2 |x y|n

K(x, y) :=
nRv(B1 (0))

for all (x, y) .


Single Second-Order PDE

Proof (cont.).
I Further, it follows for every y that
K
nRv(B1 (0)) (x, y)
xi
= 2xi |x y|n n R2 |x|2 (xi yi ) |x y|(n+2)


2K
nRv(B1 (0)) (x, y) = 2 |x y|n
xi2
+ 4nxi (xi yi ) |x y|(n+2) n R2 |x|2 |x y|(n+2)


+ n(n + 2) R2 |x|2 (xi yi )2 |x y|(n+4)



Single Second-Order PDE
Proof (cont.).
I and hence that

nRv(B1 (0)) 4K(, y) = 2n |x y|n


X n
(n+2)
xi (xi yi ) + 2n R2 |x|2 |x y|(n+2)

+ 4n |x y|
i=1
" n
#
X
= 2n |x y|n + 2n |x|2 + |y|2 2 xi yi |x y|(n+2) = 0
i=1

for all x .
I As a consequence, K(, y) is harmonic for every y .
I Hence it follows by differentiation under the integral sign
that u C 2 (, R) and that u is harmonic.
Single Second-Order PDE

Proof (cont.).
I Let x0 , > 0 and > 0 such that

|f (x) f (x0 )| <

for all x U (x0 ).


I Further, let M > 0 be such that |f (x)| 6 M for all x .
Single Second-Order PDE

Proof (cont.).
I Then it follows for every x U/2 (x0 ) that
Z

|u(x) f (x0 )| = K(x, y)(f (y) f (x0 )) dSy

Z
6 |K(x, y)| |f (y) f (x0 )| dSy
U (x0 )
Z
+ |K(x, y)| |f (y) f (x0 )| dSy
(Rn \U (x0 ))
 n Z
2M(R2 |x|2 )
6+ dSy
nRv(B1 (0)) 2
Single Second-Order PDE

Proof (cont.).
I because

|xy| = |xx0 (yx0 )| > | |xx0 ||yx0 | | > =
2 2
for all y (Rn \ U (x0 )) .
I Obviously, this implies that

lim u(x) = f (x0 ) .


xx0
Single Second-Order PDE
y

R

x 02

-R R x
W

W
-R

Figure 51: Sketch for the proof of Theorem 7.48.


Single Second-Order PDE
In the following, we study the solutions of the heat equation as
our main example for a parabolic equation.

The heat equation45 describes the temperature as a function of


position and time, T C 2 (I , R), in a homogeneous body
occupying the volume . It is given by
T
(t, x) [4T (t, )](x) = 0
t
for every (t, x) I , where (0, ) is the thermal
diffusivity46 of the body. The heat equation is linear and of
second order. It is a prototype of a so called parabolic equation.

45
This equation is also sometimes referred as diffusion equation.
46
For instance, the thermal diffusivity of iron at 27 degrees Celsius is about
0.227 cm2 /s [15].
Single Second-Order PDE

Figure 52: Sketch of P in Theorem 7.49 for the case n = 2 and


= U1 (0).
Single Second-Order PDE
Theorem 7.49
(Weak Maximum Principle) Let n N , a non-empty
bounded open subset of Rn and a < b. Further, let
u : [a, b] R be a continuous function that is continuously
partially differentiable in the first variable and twice
continuously partially differentiable in the second variable on
(a, b) as well as such that

u
(t, x) [u(t, )](x) 6 0
t
for all (t, x) (a, b) . Then u assumes its maximum value
on the part P of the boundary of [a, b] given by

P := ( {a} ) ( [a, b] ) .
Single Second-Order PDE

Proof.
I In a first step, we assume the stronger condition that
u
(t, x) [u(t, )](x) < 0 (7.54)
t
for all (t, x) (a, b) .
I Further, let N such that > 1/(b a).
I Then there is (t , x ) [a, b (1/)] such that

u(t , x ) = max u.
[a,b(1/)]
Single Second-Order PDE

Proof (cont.).
I If (t , x ) (a, b (1/)) , then

u
(t , x ) = 0 , [u(t , )](x ) 6 0
t
I and hence
u
(t , x ) [u(t , )](x ) > 0 , (7.55)
t
I in contradiction to (7.54).
Single Second-Order PDE

Proof (cont.).
I If (t , x ) {b (1/)} , then

u
(t , x ) > 0 , [u(t , )](x ) 6 0
t
I and hence (7.55) in contradiction to (7.54).
I As a consequence,

(t , x ) ( {a} ) ( [a, b (1/)] ) .


Single Second-Order PDE
Proof (cont.).
I Since P is compact, there is a strictly increasing sequence
1 , 2 , . . . in N such that ((t1 , x1 ), (t2 , x2 ), . . . ) is
convergent some (t0 , x0 ) P.
I In particular, we conclude that

u(t0 , x0 ) > u(t, x)

for every (t, x) [a, b)


I and, from the continuity of u, that

u(t0 , x0 ) > u(t, x)

for every (t, x) [a, b] .


Single Second-Order PDE
Proof (cont.).
I Hence
max u = max u .
[a,b] P

I In the following, we assume instead of (7.54) only that


u
(t, x) [u(t, )](x) 6 0
t
for all (t, x) (a, b)
I and define for > 0 a corresponding function v by
t
v (t, x) := u(t, x)

for (t, x) [a, b] .
Single Second-Order PDE

Proof (cont.).
I Then v is a continuous function that is continuously
partially differentiable in the first variable
I and twice continuously partially differentiable in the
second variable on (a, b)
I as well as such that
v u 1
(t, x) [v (t, )](x) = (t, x) [u(t, )](x) < 0
t t
for all (t, x) (a, b) .
Single Second-Order PDE

Proof (cont.).
I Hence it follows from the first part of the proof that

max v = max v .
[a,b] P

I and hence the existence of (t , x ) P such that


t t
u(t, x) = v (t, x) 6 v (t , x ) = u(t , x )

for all (t, x) [a, b] .
Single Second-Order PDE
Proof (cont.).
I The latter implies that
t t ba
u(t, x) 6 u(t , x ) + 6 u(t , x ) +

for all (t, x) [a, b] .
I Since P is compact, there is a strictly increasing sequence
1 , 2 , . . . in N such that ((t1 , x1 ), (t2 , x2 ), . . . ) is
convergent some (t0 , x0 ) P.
I In particular, we conclude, from the continuity of u, that

u(t, x) 6 u(t0 , x0 )

for all (t, x) [a, b]


Single Second-Order PDE

Proof (cont.).
I and hence that
max u = max u .
[a,b] P
Appendix
Lemma 8.1
(Transformation of the Laplace operator under inversion
with respect to n-spheres) For this, let n N , R > 0 and
Rn \ {0} be non-empty and open. In addition, let
0 Rn \ {0} be a non-empty open subset such that

g( 0 ) = ,

where g C (Rn \ {0}, Rn \ {0}) is defined by

R2
g(x) := x
|x|2

for all x Rn \ {0}. Note that g is an involution47 and that its


fixed point set is given by SRn .
47
i.e, g(g(x)) = x for all x Rn \ {0},
Appendix

Lemma (cont.)
Finally, let u C 2 (, R) and u C 2 (0 , R) be defined by
" 2n #
R2
 2 
2n R
u(x) := u (g(x)) = |x| u x
|| |x|2

for all x 0 . Then


|x|n+2
(4u)(g(x)) = (4u)(x)
R4
for all x 0 .
Appendix

Proof.
I For this, we define u C 2 (0 , R) by
 2 
R
u(x) := (u g)(x) = u x
|x|2

for all x 0 .
I Then, it follows by the chain rule that
n  2
R2

u X u R
(x) = (g(x)) 4 2xi xk ,
2 ik
xi k=1
xk |x| |x|
Appendix

Proof (cont.).
I where (
6 m
0 if l =
lm :=
1 if l = m
for all l, m {1, . . . , n},
Appendix
Proof (cont.).
2 u
(x)
xi2
n
2u
 2
R2
 2
R2

X R R
= (g(x)) ik 4 2xi xk ij 4 2xi xj
xj xk |x|2 |x| |x|2 |x|
j,k=1
n
4R2 8R2 R2
 
X u
+ (g(x)) 4 ik xi + 6 xi2 xk 4 2xk
xk |x| |x| |x|
k=1
n
R4 2 u 4R4 X 2u
= (g(x)) xi x j (g(x))
|x|4 xi2 |x|6 j=1 xi xj
n
4R4 2 X 2u
+ 8
x i xj xk (g(x))
|x| xj xk
j,k=1
n
4R2 8R2 R2
 
X u
+ (g(x)) 4 ik xi + 6 xi2 xk 4 2xk
xk |x| |x| |x|
k=1
Appendix

Proof (cont.).
I and hence
n
R4 R2 X u
(4u)(x) = (4u)(g(x)) + 2(2 n) xk (g(x))
|x|4 |x|4 k=1 xk

for every x 0 .
Appendix
Proof (cont.).
I Further,
u u
(x) = (2 n) xi |x|n u(x) + |x|2n (x) ,
xi xi
2 u
2
(x) = (2 n) |x|n u(x) n(2 n) xi2 |x|(n+2) u(x)
xi
u 2 u
+ 2(2 n) xi |x|n (x) + |x|2n 2 (x)
xi xi
n 2 (n+2)
= (2 n) |x| u(x) n(2 n) xi |x| u(x)
n
R2 R2 2
 
n
X u
+ 2(2 n) |x| (g(x)) x 4 2xi xk
2 ik i
k=1
x k |x| |x|
2 u
+ |x|2n (x)
xi2
Appendix

Proof (cont.).
I and hence
n
X u
(4u)(x) = 2(2 n) R2 |x|(n+2) xi (g(x)) + |x|2n (4u)(x)
i=1
x i
n  4
X u R
= 2(2 n) R2 |x|(n+2) xi (g(x)) + |x|2n 4
(4u)(g(x))
i=1
xi |x|
n
#
R2 X u R4
+2(2 n) 4 xk (g(x)) = n+2 (4u)(g(x))
|x| xk |x|
k=1

for every x 0 .
Appendix
Lemma 8.2
(Transformation of the Laplace operator in two dimensions
into polar coordinates) For this, let R2 \ {(0, 0)} be
non-empty and open. In addition, let p R2 be a non-empty
open subset such
g(p ) = ,
where g C (R2 , R2 ) is defined by

g(r, ) := (r cos(), r sin())

for all (r, ) R2 . Finally, let u C 2 (, R). Then

2 u 1 u 1 2 u
(4u)(g(r, )) = (r, ) + (r, ) + (r, ) (8.1)
r 2 r r r 2 2

for all (r, ) p ,


Appendix

Lemma (cont.)
where u C 2 (p , R) is defined by

u(r, ) := (u g)(r, ) = u(r cos(), r sin())

for all (r, ) p .


Appendix

Proof.
It follows by the chain rule that
u u u
(r, ) = cos() (g(r, )) + sin() (g(r, ))
r x1 x2
u u u
(r, ) = r sin() (g(r, )) + r cos() (g(r, )) ,
x1 x2
Appendix

Proof (cont.).

2 u 2u 2u
 
(r, ) = cos() cos() 2 (g(r, )) + sin() (g(r, ))
r 2 x1 x2 x1
2u 2u
 
+ sin() cos() (g(r, )) + sin() 2 (g(r, ))
x1 x2 x2
2 2
u u
= cos2 () 2 (g(r, )) + sin2 () 2 (g(r, ))
x1 x2
2
u
+ 2 sin() cos() (g(r, )) ,
x1 x2
Appendix
Proof (cont.).

2 u u
2
(r, ) = r (r, )
r
2u 2u
 
r sin() r sin() 2 (g(r, )) + r cos() (g(r, ))
x1 x2 x1
 
u u
+ r cos() r sin() (g(r, )) + r cos() 2 (g(r, ))
x1 x2 x2
2
2u

u 2 2 u 2
= r (r, ) + r sin () 2 (g(r, )) + cos () 2 (g(r, ))
r x1 x
2
 2
u
2 sin() cos() (g(r, ))
x1 x2

and hence also (8.1) for all (r, ) p .


Appendix

Lemma 8.3
(Transformation of the Laplace operator in three
dimensions into cylindrical coordinates) For this, let
R3 \ ({0} {0} R) be non-empty and open. In
addition, let cyl R3 be a non-empty open subset such

g(cyl ) = ,

where g C (R3 , R3 ) is defined by

g(r, , z) := (r cos(), r sin(), z)

for all (r, , z) R3 . Finally, let u C 2 (, R).


Appendix

Lemma (cont.)
Then
2 u 1 u 1 2 u
(4u)(g(r, , z)) = (r, , z) + (r, , z) + (r, , z)
r 2 r r r 2 2
2 u
+ 2 (r, , z) (8.2)
z
for all (r, , z) cyl , where u C 2 (cyl , R) is defined by

u(r, , z) := (u g)(r, , z) = u(r cos(), r sin(), z)

for all (r, , z) cyl .


Appendix

Proof.
It follows by the chain rule that
u u u
(r, , z) = cos() (g(r, , z)) + sin() (g(r, , z))
r x1 x2
u u u
(r, , z) = r sin() (g(r, , z)) + r cos() (g(r, , z))
x1 x2
u u
(r, , z) = (g(r, , z)) ,
z x3
Appendix
Proof (cont.).

2 u
(r, , z)
r 2
2u 2u
 
= cos() cos() 2 (g(r, , z)) + sin() (g(r, , z))
x1 x2 x1
2u 2u
 
+ sin() cos() (g(r, , z)) + sin() 2 (g(r, , z))
x1 x2 x2
2u 2u
= cos2 () 2
(g(r, , z)) + sin2 () 2 (g(r, , z))
x1 x2
2u
+ 2 sin() cos() (g(r, , z)) ,
x1 x2
Appendix
Proof (cont.).
2 u u
2
(r, , z) = r (r, , z)
r
2u 2u
 
r sin() r sin() 2 (g(r, , z)) + r cos() (g(r, , z))
x1 x2 x1
 
u u
+ r cos() r sin() (g(r, , z)) + r cos() 2 (g(r, , z))
x1 x2 x2
2 2

u u
= r 2 sin2 () 2 (g(r, , z)) + cos2 () 2 (g(r, , z))
x x2
2

u
2 sin() cos() (g(r, , z)) ,
x1 x2
2 u 2u
(r, , z) = (g(r, , z))
z2 x32

and hence also (8.2) for all (r, , z) cyl .


Appendix

Lemma 8.4
(Transformation of the Laplace operator in three
dimensions into spherical coordinates) For this, let
R3 \ ({0} {0} R) be non-empty and open. In
addition, let sph R3 be a non-empty open subset such

g(sph ) = ,

where g C (R3 , R3 ) is defined by

g(r, , ) := (r sin() cos(), r sin() sin(), r cos())

for all (r, , z) R3 . Finally, let u C 2 (, R).


Appendix
Lemma (cont.)
Then
2 u 2 u
(4u)(g(r, , z)) = 2 (r, , z) + (r, , z)
 2 r r r
1 u 2 2 u
+ 2 2 (r, , z) + sin () (r, , z)
r sin () 2 2

u
+ sin() cos() (r, , z) (8.3)

for all (r, , z) sph , where u C 2 (sph , R) is defined by

u(r, , ) := (ug)(r, , ) = u(r sin() cos(), r sin() sin(), r cos())

for all (r, , ) sph .


Appendix
Proof.
It follows by the chain rule that
u u
(r, , z) = sin() cos() (g(r, , z))
r x1
u
+ sin() sin() (g(r, , z))
x2
u
+ cos() (g(r, , z))
x3
u u
(r, , z) = r cos() cos() (g(r, , z))
x1
u
+ r cos() sin() (g(r, , z))
x2
u
r sin() (g(r, , z))
x3
Appendix

Proof (cont.).

u u
(r, , z) = r sin() sin() (g(r, , z))
x1
u
+ r sin() cos() (g(r, , z)) ,
x2
Appendix
Proof (cont.).
2 u 2 2 2u
(r, , z) = sin () cos () (g(r, , z))
r 2 x12
2u
+ sin2 () sin() cos() (g(r, , z))
x2 x1
2u
+ sin() cos() cos() (g(r, , z))
x3 x1
2u
+ sin2 () sin() cos() (g(r, , z))
x1 x2
2u
+ sin2 () sin2 () 2 (g(r, , z))
x2
2u
+ sin() cos() sin() (g(r, , z))
x3 x2
2u
+ sin() cos() cos() (g(r, , z))
x1 x3
Appendix
Proof (cont.).
2u
+ sin() cos() sin() (g(r, , z))
x2 x3
2u
+ cos2 () (g(r, , z))
x32
2u 2u
= sin2 () cos2 () (g(r, , z)) + sin 2
() sin 2
() (g(r, , z))
x12 x22
2u 2u
+ cos2 () (g(r, , z)) + 2 sin 2
() sin() cos() (g(r, , z))
x32 x1 x2
2u
+ 2 sin() cos() cos() (g(r, , z))
x1 x3
2u
+ 2 sin() cos() sin() (g(r, , z)) ,
x2 x3
Appendix
Proof (cont.).
2 u u 2 2 2 2u
(r, , z) = r (r, , z) + r cos () cos () (g(r, , z))
2 r x12
2u
+ r 2 cos2 () sin() cos() (g(r, , z))
x2 x1
2u
r 2 sin() cos() cos() (g(r, , z))
x3 x1
2u
+ r 2 cos2 () sin() cos() (g(r, , z))
x1 x2
2u
+ r 2 cos2 () sin2 () 2 (g(r, , z))
x2
2u
r 2 sin() cos() sin() (g(r, , z))
x3 x2
2u
r 2 sin() cos() cos() (g(r, , z))
x1 x3
Appendix
Proof (cont.).
2u
r 2 sin() cos() sin() (g(r, , z))
x2 x3
2u
+ r 2 sin2 () (g(r, , z))
x32
u 2u
= r (r, , z) + r 2 cos2 () cos2 () 2 (g(r, , z))
r x1
2u 2u
+ r 2 cos2 () sin2 () (g(r, , z)) + r 2
sin 2
() (g(r, , z))
x22 x32
2u
+ 2r 2 cos2 () sin() cos() (g(r, , z))
x1 x2
2u
2r 2 sin() cos() cos() (g(r, , z))
x1 x3
2u
2r 2 sin() cos() sin() (g(r, , z))
x2 x3
Appendix
Proof (cont.).
2 u u u
(r, , z) = r (r, , z) + r cos() (g(r, , z))
2 r x3
2u
+ r 2 sin2 () sin2 () 2 (g(r, , z))
x1
2u
r 2 sin2 () sin() cos() (g(r, , z))
x2 x1
2u
r 2 sin2 () sin() cos() (g(r, , z))
x1 x2
2u
+ r 2 sin2 () cos2 () 2 (g(r, , z))
x2
u u
= r (r, , z) + r cos() (g(r, , z))
r x3
2u 2u

+ r 2 sin2 () sin2 () 2 (g(r, , z)) + cos2 () 2 (g(r, , z))
x1 x2
2

u
2 sin() cos() (g(r, , z))
x1 x2
Appendix
Proof (cont.).
This implies that

2 u 1 2 u
(r, , z) + (r, , z)
r 2 r 2 2
2u 2u
= cos2 () 2 (g(r, , z)) + sin2 () 2 (g(r, , z))
x1 x2
2
u 2u
+ (g(r, , z)) + 2 sin() cos() (g(r, , z))
x32 x1 x2
1 u
(r, , z) ,
r r
Appendix
Proof (cont.).
2 u 1 cos() u
(r, , z) + 2 (r, , z)
r r r sin()
1 u cos2 () cos() u
= (r, , z) + (g(r, , z))
r r r sin() x1
cos2 () sin() u cos() u
+ (g(r, , z)) (g(r, , z))
r sin() x2 r x3
sin() cos() u sin() sin() u
+ (g(r, , z)) + (g(r, , z))
r x1 r x2
cos() u
+ (g(r, , z))
r x3
1 u cos() u sin() u
= (r, , z) + (g(r, , z)) + (g(r, , z))
r r r sin() x1 r sin() x2
1 u 1 u cos() u
= (r, , z) + 2
(r, , z) (g(r, , )) ,
r r r sin () r r sin2 () x3
Appendix

Proof (cont.).

1 2 u
(r, , z)
r 2 sin2 () 2
1 u cos() u
= 2 (r, , z) + (g(r, , z))
r sin () r r sin2 () x3
2u 2u
+ sin2 () 2 (g(r, , z)) + cos2 () 2 (g(r, , z))
x1 x2
2
u
2 sin() cos() (g(r, , z))
x1 x2
and, finally, (8.3) for all (r, , z) sph .
Index I
Partial differential equation
first-order
linear, 81
Cardanos formulas, 271
Conformal map, 534
Euclidean transformations, 528
Harmonic function, 546, 608, 617
Inversions with respect to n-spheres, 555, 676
Laplace operator, 400
Greens function, 643
Greens representation formula, 630
in cylindrical coordinates, 570, 689
in polar coordinates, 561, 684
in spherical coordinates, 574, 694
Poissons integral, 644
Index II
Poissons kernel, 657
One-point compactification of Rn , 612
Partial differential equation
a priori estimate, 178, 623
asymptotic of solutions, 123
biharmonic equation, 33
Burgers equation, 38
Hopf-Cole transformation, 38
inviscid, 234
Cauchy-Kowalevski theorem, 138
characteristic curve, 83, 195
characteristic hypersurface, 161, 392
characteristic method, 82, 193
characteristic paths, 83, 194
characteristic vector field, 82, 88, 193
characteristics, 83
Index III
conservation law, 106, 255
continuity equation, 106
definition, 18
diffusion equation, 38, 52, 664
maximum principle, 666
elliptic, 49, 405
Hopf maximum principle, 593
strong maximum principle, 586, 603
a priori estimate, 623
Dirichlet problem, 620, 628
Hadamards example, 516
initial boundary value problem, 516
energy method, 167
Euler, 234
first-order
quasi-linear, 192
Index IV
semi-linear, 199
Hamilton-Jacobi, 20
heat equation, 38, 52, 57, 429, 664
maximum principle, 666
hyperbolic, 53, 405
Klein-Gordon, 400, 469
Laplaces equation, 33, 55, 56
linear, 28
mass conservation, 106
method of descent, 459
method of separation of variables, 509
Navier-Stokes equation, 38
Hopf-Cole transformation, 38
non-characteristic hypersurface, 161, 392
non-linear, 28
nonuniqueness of solutions, 141, 147
Index V
order, 19
parabolic, 52, 405, 664
Poissons equation, 49, 620
principal part, 82, 194
quantum field theory, 33
second-order
conservation laws, 485, 493
linear, constant coefficients, 419
quasi-linear, 388
shock condition, 261
shock solution, 249
singular point, 103
symmetry transformation, 480
telegraph equation, 33
transformations, 45, 420, 427, 528, 560
Tricomi, 409
Index VI
ultrahyperbolic, 405
uniqueness of the solutions, 180, 504
wave equation
boundary value problem, 506
focusing effect, 60, 455
backward characteristic cone, 404
caustic, 455
conservation laws, 475
domain of dependence, 373, 450, 464, 474
domain of influence, 374, 452, 465, 504
Duhamels integral, 472
energy density, 486, 494
energy flux density, 486, 494
energy inequality, 498
forward characteristic cone, 404
Huygens principle, 449, 464
Index VII
in one space dimension, 43, 362
in three space dimensions, 53, 59, 430
in two space dimensions, 58, 458
inhomogeneous, 362, 471
inhomogeneous solution, 368, 473
initial value problem is well-posed, 375, 450, 453, 466
instability, 381
Kirchhoffs formula, 441
local uniqueness of the solutions, 504
solution, 368, 441, 463
weak solution, 257
Separation constants, 513
Special Relativity, 452
Stereographic projection, 612
System of PDE
Cauchy-Riemann, 29, 292, 538
Index VIII
characteristic hypersurface, 287
definition, 27
Euler, 316
first-order
elliptic, 341
hyperbolic, 341
quasi-linear, 283
symmetric hyperbolic, 341
Maxwells equations, 33
method of separation of variables, 336
plane waves, 336
Riemann invariants, 332
symmetric hyperbolic, 283, 303
dissipative boundary condition, 310
uniqueness of the solutions, 303
vector fields
Index IX

direction field, 67
integral curves, 75
maximal, 77
References I
[1] Bers L, John F, Schechter M 1964, Partial differential
equations, Wiley: New York.
[2] Beyer H R 2010, Calculus and Analysis: A Combined
Approach, Wiley: New York.
[3] Beyer H R 2007, Beyond partial differential equations: A
course on linear and quasi-linear abstract hyperbolic
evolution equations, Springer Lecture Notes in
Mathematics 1898, Springer: Berlin.
[4] Coddington, E A 1989, An introduction to ordinary
differential equations, Dover : New York.
[5] CODATA (Committee on Data for Science and
Technology), 2006.
[6] Coddington, E A, Levinson, N 1955, Theory of ordinary
differential equations, Dover : New York.
References II
[7] DiBenedetto E 1995, Partial differential equations,
Birkhuser: Boston.
[8] Evans L C 2002, Partial differential equations, AMS:
Providence.
[9] Garabedian P R 1988, Partial differential equations, AMS:
Providence.
[10] Gilbarg D, Trudinger N S 1977, Elliptic partial differential
equations of second order, Springer: Berlin.
[11] Hille, E A 1969, Lectures on ordinary differential
equations, Addison-Wesley : Reading,Mass.
[12] John F 1982, Partial differential equations, 4th ed.,
Springer: New York.
[13] Jost J 2002, Partial differential equations, Springer: New
York.
References III
[14] Kaye G W C, Laby T H 1995, Tables of Physical and
Chemical Constants: And Some Mathematical Functions,
16th ed., Longman Scientific and Technical, London.
[15] Monaghan B J, Quested P N 2001, Thermal diffusivity of
iron at high temperature in both the liquid and solid states,
ISIJ International, 41, 1524-1528.
[16] Morawetz, C S 1975, Notes on time decay and scattering
for some hyperbolic problems, Society for Industrial
Mathematics: Philadelphia.
[17] Protter M H and Weinberger H F 1984, Maximum
principles in differential equations, Springer: New York.
[18] Rauch J 1991, Partial differential equations, Springer:
New York.
[19] Renardy M and Rogers R C 1993, An introduction to
partial differential equations, Springer: New York.
References IV

[20] Smirnov M M 1966, Second-order partial differential


equations, Noordhoff: Groningen.
[21] Spivak M 1979, A comprehensive introduction to
differential geometry, Vol. V, Publish or Perish: Berkeley.
[22] Walter, W 1998, Ordinary differential equations, Springer
: New York.
[23] Zachmanoglou E C, Thoe D W 1986, Introduction to
partial differential equations with applications, Dover:
New York.

Vous aimerez peut-être aussi