Pde Book

Partial Dierential Equations
Lecture Notes
Erich Miersemann
Department of Mathematics
Leipzig University
Version October, 2012
2
Contents
1 Introduction 9
1.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Equations from variational problems . . . . . . . . . . . . . . 15
1.2.1 Ordinary dierential equations . . . . . . . . . . . . . 15
1.2.2 Partial dierential equations . . . . . . . . . . . . . . 16
1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2 Equations of rst order 25
2.1 Linear equations . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Quasilinear equations . . . . . . . . . . . . . . . . . . . . . . 31
2.2.1 A linearization method . . . . . . . . . . . . . . . . . 32
2.2.2 Initial value problem of Cauchy . . . . . . . . . . . . . 33
2.3 Nonlinear equations in two variables . . . . . . . . . . . . . . 40
2.3.1 Initial value problem of Cauchy . . . . . . . . . . . . . 48
2.4 Nonlinear equations in R
n
. . . . . . . . . . . . . . . . . . . . 51
2.5 Hamilton-Jacobi theory . . . . . . . . . . . . . . . . . . . . . 53
2.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3 Classication 63
3.1 Linear equations of second order . . . . . . . . . . . . . . . . 63
3.1.1 Normal form in two variables . . . . . . . . . . . . . . 69
3.2 Quasilinear equations of second order . . . . . . . . . . . . . . 73
3.2.1 Quasilinear elliptic equations . . . . . . . . . . . . . . 73
3.3 Systems of rst order . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.4 Systems of second order . . . . . . . . . . . . . . . . . . . . . 82
3.4.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5 Theorem of Cauchy-Kovalevskaya . . . . . . . . . . . . . . . . 84
3.5.1 Appendix: Real analytic functions . . . . . . . . . . . 90
3
4 CONTENTS
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4 Hyperbolic equations 107
4.1 One-dimensional wave equation . . . . . . . . . . . . . . . . . 107
4.2 Higher dimensions . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2.1 Case n=3 . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2.2 Case n = 2 . . . . . . . . . . . . . . . . . . . . . . . . 115
4.3 Inhomogeneous equation . . . . . . . . . . . . . . . . . . . . . 117
4.4 A method of Riemann . . . . . . . . . . . . . . . . . . . . . . 120
4.5 Initial-boundary value problems . . . . . . . . . . . . . . . . . 125
4.5.1 Oscillation of a string . . . . . . . . . . . . . . . . . . 125
4.5.2 Oscillation of a membrane . . . . . . . . . . . . . . . . 128
4.5.3 Inhomogeneous wave equations . . . . . . . . . . . . . 131
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5 Fourier transform 141
5.1 Denition, properties . . . . . . . . . . . . . . . . . . . . . . . 141
5.1.1 Pseudodierential operators . . . . . . . . . . . . . . . 146
5.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6 Parabolic equations 151
6.1 Poissons formula . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.2 Inhomogeneous heat equation . . . . . . . . . . . . . . . . . . 155
6.3 Maximum principle . . . . . . . . . . . . . . . . . . . . . . . . 156
6.4 Initial-boundary value problem . . . . . . . . . . . . . . . . . 162
6.4.1 Fouriers method . . . . . . . . . . . . . . . . . . . . . 162
6.4.2 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . 164
6.5 Black-Scholes equation . . . . . . . . . . . . . . . . . . . . . . 164
6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
7 Elliptic equations of second order 175
7.1 Fundamental solution . . . . . . . . . . . . . . . . . . . . . . 175
7.2 Representation formula . . . . . . . . . . . . . . . . . . . . . 177
7.2.1 Conclusions from the representation formula . . . . . 179
7.3 Boundary value problems . . . . . . . . . . . . . . . . . . . . 181
7.3.1 Dirichlet problem . . . . . . . . . . . . . . . . . . . . . 181
7.3.2 Neumann problem . . . . . . . . . . . . . . . . . . . . 182
7.3.3 Mixed boundary value problem . . . . . . . . . . . . . 183
7.4 Greens function for . . . . . . . . . . . . . . . . . . . . . . 183
7.4.1 Greens function for a ball . . . . . . . . . . . . . . . . 186
CONTENTS 5
7.4.2 Greens function and conformal mapping . . . . . . . 190
7.5 Inhomogeneous equation . . . . . . . . . . . . . . . . . . . . . 190
7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
6 CONTENTS
Preface
These lecture notes are intented as a straightforward introduction to partial
dierential equations which can serve as a textbook for undergraduate and
beginning graduate students.
For additional reading we recommend following books: W. I. Smirnov [21],
I. G. Petrowski [17], P. R. Garabedian [8], W. A. Strauss [23], F. John [10],
L. C. Evans [5] and R. Courant and D. Hilbert[4] and D. Gilbarg and N. S.
Trudinger [9]. Some material of these lecture notes was taken from some of
these books.
7
8 CONTENTS
Chapter 1
Introduction
Ordinary and partial dierential equations occur in many applications. An
ordinary dierential equation is a special case of a partial dierential equa-
tion but the behaviour of solutions is quite dierent in general. It is much
more complicated in the case of partial dierential equations caused by the
fact that the functions for which we are looking at are functions of more
than one independent variable.
Equation
F(x, y(x), y
(x), . . . , y
(n)
) = 0
is an ordinary dierential equation of n-th order for the unknown function
y(x), where F is given.
An important problem for ordinary dierential equations is the initial
value problem
y
(x) = f(x, y(x))

y(x
0
) = y
0
,
where f is a given real function of two variables x, y and x
0
, y
0
are given
real numbers.
Picard-Lindel of Theorem. Suppose
(i) f(x, y) is continuous in a rectangle
Q = (x, y) R
2
: [x x
0
[ < a, [y y
0
[ < b.
(ii) There is a constant K such that [f(x, y)[ K for all (x, y) Q.
(ii) Lipschitz condition: There is a constant L such that
[f(x, y
2
) f(x, y
1
)[ L[y
2
y
1
[
9
10 CHAPTER 1. INTRODUCTION
x
y
x
y
0
0
Figure 1.1: Initial value problem
for all (x, y
1
), (x, y
2
).
Then there exists a unique solution y C
1
(x
0
, x
0
+) of the above initial
value problem, where = min(b/K, a).
The linear ordinary dierential equation
y
(n)
+a
n1
(x)y
(n1)
+. . . a
1
(x)y
+a
0
(x)y = 0,
where a
j
are continuous functions, has exactly n linearly independent solu-
tions. In contrast to this property the partial dierential u
xx
+u
yy
= 0 in R
2
has innitely many linearly independent solutions in the linear space C
2
(R
2
).
The ordinary dierential equation of second order
y
(x) = f(x, y(x), y
(x))
has in general a family of solutions with two free parameters. Thus, it is
naturally to consider the associated initial value problem
y
(x) = f(x, y(x), y
(x))
y(x
0
) = y
0
, y
(x
0
) = y
1
,
where y
0
and y
1
are given, or to consider the boundary value problem
y
(x) = f(x, y(x), y
(x))
y(x
0
) = y
0
, y(x
1
) = y
1
.
Initial and boundary value problems play an important role also in the
theory of partial dierential equations. A partial dierential equation for
1.1. EXAMPLES 11
y
y
0
x x
y
1
0
x
1
Figure 1.2: Boundary value problem
the unknown function u(x, y) is for example
F(x, y, u, u
x
, u
y
, u
xx
, u
xy
, u
yy
) = 0,
where the function F is given. This equation is of second order.
An equation is said to be of n-th order if the highest derivative which
occurs is of order n.
An equation is said to be linear if the unknown function and its deriva-
tives are linear in F. For example,
a(x, y)u
x
+b(x, y)u
y
+c(x, y)u = f(x, y),
where the functions a, b, c and f are given, is a linear equation of rst
order.
An equation is said to be quasilinear if it is linear in the highest deriva-
tives. For example,
a(x, y, u, u
x
, u
y
)u
xx
+b(x, y, u, u
x
, u
y
)u
xy
+c(x, y, u, u
x
, u
y
)u
yy
= 0
is a quasilinear equation of second order.
1.1 Examples
1. u
y
= 0, where u = u(x, y). All functions u = w(x) are solutions.
2. u
x
= u
y
, where u = u(x, y). A change of coordinates transforms this
equation into an equation of the rst example. Set = x + y, = x y,
then
u(x, y) = u
_
+
2
,

2
_
=: v(, ).
Assume u C
1
, then
v
=
1
2
(u
x
u
y
).
If u
x
= u
y
, then v
= 0 and vice versa, thus v = w() are solutions for

arbitrary C
1
-functions w(). Consequently, we have a large class of solutions
of the original partial dierential equation: u = w(x + y) with an arbitrary
C
1
-function w.
3. A necessary and sucient condition such that for given C
1
-functions
M, N the integral
_
P
1
P
0
M(x, y)dx +N(x, y)dy
is independent of the curve which connects the points P
0
with P
1
in a simply
connected domain R
2
is the partial dierential equation (condition of
integrability)
M
y
= N
x
in .
y
x
P
P
0
1
Figure 1.3: Independence of the path
This is one equation for two functions. A large class of solutions is given
by M =
x
, N =
y
, where (x, y) is an arbitrary C
2
-function. It follows
from Gauss theorem that these are all C
1
-solutions of the above dierential
equation.
4. Method of an integrating multiplier for an ordinary dierential equation.
Consider the ordinary dierential equation
M(x, y)dx +N(x, y)dy = 0
1.1. EXAMPLES 13
for given C
1
-functions M, N. Then we seek a C
1
-function (x, y) such that
Mdx +Ndy is a total dierential, i. e., that (M)
y
= (N)
x
is satised.
This is a linear partial dierential equation of rst order for :
M
y
N
x
= (N
x
M
y
).
5. Two C
1
-functions u(x, y) and v(x, y) are said to be functionally dependent
if
det
_
u
x
u
y
v
x
v
y
_
= 0,
which is a linear partial dierential equation of rst order for u if v is a given
C
1
-function. A large class of solutions is given by
u = H(v(x, y)),
where H is an arbitrary C
1
-function.
6. Cauchy-Riemann equations. Set f(z) = u(x, y)+iv(x, y), where z = x+iy
and u, v are given C
1
()-functions. Here is a domain in R
2
. If the function
f(z) is dierentiable with respect to the complex variable z then u, v satisfy
the Cauchy-Riemann equations
u
x
= v
y
, u
y
= v
x
.
It is known from the theory of functions of one complex variable that the
real part u and the imaginary part v of a dierentiable function f(z) are
solutions of the Laplace equation
u = 0, v = 0,
where u = u
xx
+u
yy
.
7. The Newton potential
u =
1
_
x
2
+y
2
+z
2
is a solution of the Laplace equation in R
3
(0, 0, 0), i. e., of
u
xx
+u
yy
+u
zz
= 0.
8. Heat equation. Let u(x, t) be the temperature of a point x at time
t, where R
3
is a domain. Then u(x, t) satises in [0, ) the heat
equation
u
t
= ku,
where u = u
x
1
x
1
+u
x
2
x
2
+u
x
3
x
3
and k is a positive constant. The condition
u(x, 0) = u
0
(x), x ,
where u
0
(x) is given, is an initial condition associated to the above heat
equation. The condition
u(x, t) = h(x, t), x , t 0,
where h(x, t) is given is a boundary condition for the heat equation.
If h(x, t) = g(x), that is, h is independent of t, then one expects that the
solution u(x, t) tends to a function v(x) if t . Moreover, it turns out
that v is the solution of the boundary value problem for the Laplace equation
v = 0 in
v = g(x) on .
9. Wave equation. The wave equation
y
u(x,t )
u(x,t ) 1
2
x
l
Figure 1.4: Oscillating string
u
tt
= c
2
u,
where u = u(x, t), c is a positive constant, describes oscillations of mem-
branes or of three dimensional domains, for example. In the one-dimensional
case
u
tt
= c
2
u
xx
describes oscillations of a string.
1.2. EQUATIONS FROM VARIATIONAL PROBLEMS 15
Associated initial conditions are
u(x, 0) = u
0
(x), u
t
(x, 0) = u
1
(x),
where u
0
, u
1
are given functions. Thus the initial position and the initial
velocity are prescribed.
If the string is nite one describes additionally boundary conditions, for
example
u(0, t) = 0, u(l, t) = 0 for all t 0.
1.2 Equations from variational problems
A large class of ordinary and partial dierential equations arise from varia-
tional problems.
1.2.1 Ordinary dierential equations
Set
E(v) =
_
b
a
f(x, v(x), v
(x)) dx
and for given u
a
, u
b
R
V = v C
2
[a, b] : v(a) = u
a
, v(b) = u
b
,
where < a < b < and f is suciently regular. One of the basic
problems in the calculus of variation is
(P) min
vV
E(v).
Euler equation. Let u V be a solution of (P), then
d
dx
f
u
(x, u(x), u
(x)) = f
u
(x, u(x), u
(x))
in (a, b).
Proof. Exercise. Hints: For xed C
2
[a, b] with (a) = (b) = 0 and
real , [[ <
0
, set g() = E(u + ). Since g(0) g() it follows g
(0) = 0.
Integration by parts in the formula for g
(0) and the following basic lemma

in the calculus of variations imply Eulers equation.
y
y
0
y
1
x a
b
Figure 1.5: Admissible variations
Basic lemma in the calculus of variations. Let h C(a, b) and
_
b
a
h(x)(x) dx = 0
for all C
1
0
(a, b). Then h(x) 0 on (a, b).
Proof. Assume h(x
0
) > 0 for an x
0
(a, b), then there is a > 0 such that
(x
0
, x
0
+) (a, b) and h(x) h(x
0
)/2 on (x
0
, x
0
+). Set
(x) =
_ _
2
[x x
0
[
2
_
2
if x (x
0
, x
0
+)
0 if x (a, b) [x
0
, x
0
+]
.
Thus C
1
0
(a, b) and
_
b
a
h(x)(x) dx
h(x
0
)
2
_
x
0
+
x
0
(x) dx > 0,
which is a contradiction to the assumption of the lemma. 2
1.2.2 Partial dierential equations
The same procedure as above applied to the following multiple integral leads
to a second-order quasilinear partial dierential equation. Set
E(v) =
_
F(x, v, v) dx,
where R
n
is a domain, x = (x
1
, . . . , x
n
), v = v(x) : R, and
v = (v
x
1
, . . . , v
x
n
). Assume that the function F is suciently regular in
its arguments. For a given function h, dened on , set
V = v C
2
() : v = h on .
Euler equation. Let u V be a solution of (P), then
n
i=1
x
i
F
u
x
i
F
u
= 0
in .
Proof. Exercise. Hint: Extend the above fundamental lemma of the calculus
of variations to the case of multiple integrals. The interval (x
0
, x
0
+) in
the denition of must be replaced by a ball with center at x
0
and radius
.
Example: Dirichlet integral
In two dimensions the Dirichlet integral is given by
D(v) =
_
_
v
2
x
+v
2
y
_
dxdy
and the associated Euler equation is the Laplace equation u = 0 in .
Thus, there is natural relationship between the boundary value problem
u = 0 in , u = h on
and the variational problem
min
vV
D(v).
But these problems are not equivalent in general. It can happen that the
boundary value problem has a solution but the variational problem has no
solution, see for an example Courant and Hilbert [4], Vol. 1, p. 155, where
h is a continuous function and the associated solution u of the boundary
value problem has no nite Dirichlet integral.
The problems are equivalent, provided the given boundary value function
h is in the class H
1/2
(), see Lions and Magenes [14].
Example: Minimal surface equation
The non-parametric minimal surface problem in two dimensions is to
nd a minimizer u = u(x
1
, x
2
) of the problem
min
vV
_
_
1 +v
2
x
1
+v
2
x
2
dx,
where for a given function h dened on the boundary of the domain
V = v C
1
() : v = h on .
S
Figure 1.6: Comparison surface

Suppose that the minimizer satises the regularity assumption u C
2
(),
then u is a solution of the minimal surface equation (Euler equation) in
x
1
_
u
x
1
_
1 +[u[
2
_
+

x
2
_
u
x
2
_
1 +[u[
2
_
= 0. (1.1)
In fact, the additional assumption u C
2
() is superuous since it follows
from regularity considerations for quasilinear elliptic equations of second
order, see for example Gilbarg and Trudinger [9].
Let = R
2
. Each linear function is a solution of the minimal surface
equation (1.1). It was shown by Bernstein [2] that there are no other solu-
tions of the minimal surface quation. This is true also for higher dimensions
n 7, see Simons [19]. If n 8, then there exists also other solutions which
dene cones, see Bombieri, De Giorgi and Giusti [3].
The linearized minimal surface equation over u 0 is the Laplace equa-
tion u = 0. In R
2
linear functions are solutions but also many other
functions in contrast to the minimal surface equation. This striking dier-
ence is caused by the strong nonlinearity of the minimal surface equation.
More general minimal surfaces are described by using parametric rep-
resentations. An example is shown in Figure 1.7
1
. See [18], pp. 62, for
example, for rotationally symmetric minimal surfaces.
Figure 1.7: Rotationally symmetric minimal surface
Neumann type boundary value problems
Set V = C
1
() and
E(v) =
_
F(x, v, v) dx
_
g(x, v) ds,
where F and g are given suciently regular functions and R
n
is a
bounded and suciently regular domain. Assume u is a minimizer of E(v)
in V , that is
u V : E(u) E(v) for all v V,
1
An experiment from Beutelspachers Mathematikum, Wissenschaftsjahr 2008, Leipzig
then
_
_
n
i=1
F
u
x
i
(x, u, u)
x
i
+ F
u
(x, u, u)
_
dx
g
u
(x, u) ds = 0
for all C
1
(). Assume additionally u C
2
(), then u is a solution of
the Neumann type boundary value problem
n
i=1
x
i
F
u
x
i
F
u
= 0 in
n
i=1
F
u
x
i
i
g
u
= 0 on ,
where = (
1
, . . . ,
n
) is the exterior unit normal at the boundary . This
follows after integration by parts from the basic lemma of the calculus of
variations.
Example: Laplace equation
Set
E(v) =
1
2
_
[v[
2
dx
_
h(x)v ds,
then the associated boundary value problem is
u = 0 in
u
= h on .
Example: Capillary equation
Let R
2
and set
E(v) =
_
_
1 +[v[
2
dx +

2
_
v
2
dx cos
_
v ds.
Here is a positive constant (capillarity constant) and is the (constant)
boundary contact angle, i. e., the angle between the container wall and
the capillary surface, dened by v = v(x
1
, x
2
), at the boundary. Then the
related boundary value problem is
div (Tu) = u in
Tu = cos on ,
where we use the abbreviation
Tu =
u
_
1 +[u[
2
,
div (Tu) is the left hand side of the minimal surface equation (1.1) and it
is twice the mean curvature of the surface dened by z = u(x
1
, x
2
), see an
exercise.
The above problem describes the ascent of a liquid, water for example,
in a vertical cylinder with cross section . Assume the gravity is directed
downwards in the direction of the negative x
3
-axis. Figure 1.8 shows that
liquid can rise along a vertical wedge which is a consequence of the strong
nonlinearity of the underlying equations, see Finn [7]. This photo was taken
Figure 1.8: Ascent of liquid in a wedge
from [15].
1.3 Exercises
1. Find nontrivial solutions u of
u
x
y u
y
x = 0 .
2. Prove: In the linear space C
2
(R
2
) there are innitely many linearly
independent solutions of u = 0 in R
2
.
Hint: Real and imaginary part of holomorphic functions are solutions
of the Laplace equation.
3. Find all radially symmetric functions which satisfy the Laplace equa-
tion in R
n
0 for n 2. A function u is said to be radially symmetric
if u(x) = f(r), where r = (
n
i
x
2
i
)
1/2
.
Hint: Show that a radially symmetric u satises u = r
1n
_
r
n1
f
by using u(x) = f
(r)
x
r
.
4. Prove the basic lemma in the calculus of variations: Let R
n
be a
domain and f C() such that
_
f(x)h(x) dx = 0
for all h C
2
0
(). Then f 0 in .
5. Write the minimal surface equation (1.1) as a quasilinear equation of
second order.
6. Prove that a suciently regular minimizer in C
1
() of
E(v) =
_
F(x, v, v) dx
_
g(v, v) ds,
is a solution of the boundary value problem
n
i=1
x
i
F
u
x
i
F
u
= 0 in
n
i=1
F
u
x
i
i
g
u
= 0 on ,
where = (
1
, . . . ,
n
) is the exterior unit normal at the boundary .
1.3. EXERCISES 23
7. Prove that Tu = cos on , where is the angle between the
container wall, which is here a cylinder, and the surface S, dened by
z = u(x
1
, x
2
), at the boundary of S, is the exterior normal at .
Hint: The angle between two surfaces is by denition the angle between
the two associated normals at the intersection of the surfaces.
8. Let be bounded and assume u C
2
() is a solution of
div Tu = C in

u
_
1 +[u[
2
= cos on ,
where C is a constant.
Prove that
C =
[[
[[
cos .
Hint: Integrate the dierential equation over .
9. Assume = B
R
(0) is a disc with radius R and the center at the origin.
Show that radially symmetric solutions u(x) = w(r), r =
_
x
2
1
+x
2
2
,
of the capillary boundary value problem are solutions of
_
rw
1 +w
2
_
= rw in 0 < r < R
w
1 +w
2
= cos if r = R.
Remark. It follows from a maximum principle of Concus and Finn [7]
that a solution of the capillary equation over a disc must be radially
symmetric.
10. Find all radially symmetric solutions of
_
rw
1 +w
2
_
= Cr in 0 < r < R
w
1 +w
2
= cos if r = R.
Hint: From an exercise above it follows that
C =
2
R
cos .
11. Show that div Tu is twice the mean curvature of the surface dened
by z = u(x
1
, x
2
).
Chapter 2
Equations of rst order
For a given suciently regular function F the general equation of rst order
for the unknown function u(x) is
F(x, u, u) = 0
in R
n
. The main tool for studying related problems is the theory of
ordinary dierential equations. This is quite dierent for systems of partial
dierential of rst order.
The general linear partial dierential equation of rst order can be writ-
ten as
n
i=1
a
i
(x)u
x
i
+c(x)u = f(x)
for given functions a
i
, c and f. The general quasilinear partial dierential
equation of rst order is
n
i=1
a
i
(x, u)u
x
i
+c(x, u) = 0.
2.1 Linear equations
Let us begin with the linear homogeneous equation
a
1
(x, y)u
x
+a
2
(x, y)u
y
= 0. (2.1)
Assume there is a C
1
-solution z = u(x, y). This function denes a surface
S which has at P = (x, y, u(x, y)) the normal
N =
1
_
1 +[u[
2
(u
x
, u
y
, 1)
25
26 CHAPTER 2. EQUATIONS OF FIRST ORDER
and the tangential plane dened by
z = u
x
(x, y)( x) +u
y
(x, y)( y).
Set p = u
x
(x, y), q = u
y
(x, y) and z = u(x, y). The tuple (x, y, z, p, q) is
called surface element and the tuple (x, y, z) support of the surface element.
The tangential plane is dened by the surface element. On the other hand,
dierential equation (2.1)
a
1
(x, y)p +a
2
(x, y)q = 0
denes at each support (x, y, z) a bundle of planes if we consider all (p, q) sat-
isfying this equation. For xed (x, y), this family of planes () = (; x, y)
is dened by a one parameter family of ascents p() = p(; x, y), q() =
q(; x, y). The envelope of these planes is a line since
a
1
(x, y)p() +a
2
(x, y)q() = 0,
which implies that the normal N() on () is perpendicular on (a
1
, a
2
, 0).
Consider a curve x() = (x(), y(), z()) on o, let T
x
0
be the tangential
plane at x
0
= (x(
0
), y(
0
), z(
0
)) of o and consider on T
x
0
the line
L : l() = x
0
+x
(
0
), R,
see Figure 2.1.
We assume L coincides with the envelope, which is a line here, of the
family of planes () at (x, y, z). Assume that T
x
0
= (
0
) and consider
two planes
(
0
) : z z
0
= (x x
0
)p(
0
) + (y y
0
)q(
0
)
(
0
+h) : z z
0
= (x x
0
)p(
0
+h) + (y y
0
)q(
0
+h).
At the intersection l() we have
(x x
0
)p(
0
) + (y y
0
)q(
0
) = (x x
0
)p(
0
+h) + (y y
0
)q(
0
+h).
Thus,
x
(
0
)p
(
0
) +y
(
0
)q
(
0
) = 0.
From the dierential equation
a
1
(x(
0
), y(
0
))p() +a
2
(x(
0
), y(
0
))q() = 0
2.1. LINEAR EQUATIONS 27
y
z
x
L
S
(
0
)
Figure 2.1: Curve on a surface
it follows
a
1
p
(
0
) +a
2
q
(
0
) = 0.
Consequently
(x
(), y
()) =
x
()
a
1
(x(, y())
(a
1
(x(), y()), a
2
(x(), y()),
since
0
was an arbitrary parameter. Here we assume that x
() ,= 0 and
a
1
(x(), y()) ,= 0.
Then we introduce a new parameter t by the inverse of = (t), where
t() =
_

0
x
(s)
a
1
(x(s), y(s))
ds.
It follows x
(t) = a
1
(x, y), y
(t) = a
2
(x, y). We denote x((t)) by x(t) again.
Now we consider the initial value problem
x
(t) = a
1
(x, y), y
(t) = a
2
(x, y), x(0) = x
0
, y(0) = y
0
. (2.2)
From the theory of ordinary dierential equations it follows (Theorem of
Picard-Lindelof) that there is a unique solution in a neighbouhood of t = 0
provided the functions a
1
, a
2
are in C
1
. From this denition of the curves
(x(t), y(t)) is follows that the eld of directions (a
1
(x
0
, y
0
), a
2
(x
0
, y
0
)) denes
the slope of these curves at (x(0), y(0)).
Denition. The dierential equations in (2.2) are called characteristic
equations or characteristic system and solutions of the associated initial value
problem are called characteristic curves.
Denition. A function (x, y) is said to be an integral of the characteristic
system if (x(t), y(t)) = const. for each characteristic curve. The constant
depends on the characteristic curve considered.
Proposition 2.1. Assume C
1
is an integral, then u = (x, y) is a
solution of (2.1).
Proof. Consider for given (x
0
, y
0
) the above initial value problem (2.2).
Since (x(t), y(t)) = const. it follows
x
x
+
y
y
= 0
for [t[ < t
0
, t
0
> 0 and suciently small. Thus
x
(x
0
, y
0
)a
1
(x
0
, y
0
) +
y
(x
0
, y
0
)a
2
(x
0
, y
0
) = 0.
2
Remark. If (x, y) is a solution of equation (2.1) then also H((x, y)),
where H(s) is a given C
1
-function.
Examples
1. Consider
a
1
u
x
+a
2
u
y
= 0,
where a
1
, a
2
are constants. The system of characteristic equations is
x
= a
1
, y
= a
2
.
Thus the characteristic curves are parallel straight lines dened by
x = a
1
t +A, y = a
2
t +B,
2.1. LINEAR EQUATIONS 29
where A, B are arbitrary constants. From these equations it follows that
(x, y) := a
2
x a
1
y
is constant along each characteristic curve. Consequently, see Proposi-
tion 2.1, u = a
2
x a
1
y is a solution of the dierential equation. From
an exercise it follows that
u = H(a
2
x a
1
y), (2.3)
where H(s) is an arbitrary C
1
-function, is also a solution. Since u is constant
when a
2
x a
1
y is constant, equation (2.3) denes cylinder surfaces which
are generated by parallel straight lines which are parallel to the (x, y)-plane,
see Figure 2.2.
y
z
x
Figure 2.2: Cylinder surfaces
2. Consider the dierential equation
xu
x
+yu
y
= 0.
The characteristic equations are
x
= x, y
= y,
and the characteristic curves are given by
x = Ae
t
, y = Be
t
,
where A, B are arbitrary constants. Thus, an integral is y/x, x ,= 0, and for
a given C
1
-function the function u = H(x/y) is a solution of the dierential
equation. If y/x = const., then u is constant. Suppose that H
(s) > 0,
for example, then u denes right helicoids (in German: Wendelachen), see
Figure 2.3
Figure 2.3: Right helicoid, a
2
< x
2
+ y
2
< R
2
(Museo Ideale Leonardo da
Vinci, Italy)
3. Consider the dierential equation
yu
x
xu
y
= 0.
The associated characteristic system is
x
= y, y
= x.
If follows
x
x +yy
= 0,
2.2. QUASILINEAR EQUATIONS 31
or, equivalently,
d
dt
(x
2
+y
2
) = 0,
which implies that x
2
+ y
2
= const. along each characteristic. Thus, rota-
tionally symmetric surfaces dened by u = H(x
2
+ y
2
), where H
,= 0, are
solutions of the dierential equation.
4. The associated characteristic equations to
ayu
x
+bxu
y
= 0,
where a, b are positive constants, are given by
x
= ay, y
= bx.
It follows bxx
ayy
= 0, or equivalently,
d
dt
(bx
2
ay
2
) = 0.
Solutions of the dierential equation are u = H(bx
2
ay
2
), which dene
surfaces which have a hyperbola as the intersection with planes parallel to
the (x, y)-plane. Here H(s) is an arbitrary C
1
-function, H
(s) ,= 0.
2.2 Quasilinear equations
Here we consider the equation
a
1
(x, y, u)u
x
+a
2
(x, y, u)u
y
= a
3
(x, y, u). (2.4)
The inhomogeneous linear equation
a
1
(x, y)u
x
+a
2
(x, y)u
y
= a
3
(x, y)
is a special case of (2.4).
One arrives at characteristic equations x
= a
1
, y
= a
2
, z
= a
3
from (2.4) by the same arguments as in the case of homogeneous linear
equations in two variables. The additional equation z
= a
3
follows from
z
() = p()x
() +q()y
()
= pa
1
+qa
2
= a
3
,
see also Section 2.3, where the general case of nonlinear equations in two
variables is considered.
2.2.1 A linearization method
We can transform the inhomogeneous equation (2.4) into a homogeneous
linear equation for an unknown function of three variables by the following
trick.
We are looking for a function (x, y, u) such that the solution u = u(x, y)
of (2.4) is dened implicitly by (x, y, u) = const. Assume there is such a
function and let u be a solution of (2.4), then
x
+
u
u
x
= 0,
y
+
u
u
y
= 0.
Assume
u
,= 0, then
u
x
=
u
, u
y
=
u
.
From (2.4) we obtain
a
1
(x, y, z)
x
+a
2
(x, y, z)
y
+a
3
(x, y, z)
z
= 0, (2.5)
where z := u.
We consider the associated system of characteristic equations
x
(t) = a
1
(x, y, z)
y
(t) = a
2
(x, y, z)
z
(t) = a
3
(x, y, z).
One arrives at this system by the same arguments as in the two-dimensional
case above.
Proposition 2.2. (i) Assume w C
1
, w = w(x, y, z), is an integral, i.
e., it is constant along each xed solution of (2.5), then = w(x, y, z) is a
solution of (2.5).
(ii) The function z = u(x, y), implicitly dened through (x, u, z) = const.,
is a solution of (2.4), provided that
z
,= 0.
(iii) Let z = u(x, y) be a solution of (2.4) and let (x(t), y(t)) be a solution of
x
(t) = a
1
(x, y, u(x, y)), y
(t) = a
2
(x, y, u(x, y)),
then z(t) := u(x(t), y(t)) satises the third of the above characteristic equa-
tions.
Proof. Exercise.
2.2.2 Initial value problem of Cauchy
Consider again the quasilinear equation
() a
1
(x, y, u)u
x
+a
2
(x, y, u)u
y
= a
3
(x, y, u).
Let
: x = x
0
(s), y = y
0
(s), z = z
0
(s), s
1
s s
2
, < s
1
< s
2
< +
be a regular curve in R
3
and denote by ( the orthogonal projection of
onto the (x, y)-plane, i. e.,
( : x = x
0
(s), y = y
0
(s).
Initial value problem of Cauchy: Find a C
1
-solution u = u(x, y) of
() such that u(x
0
(s), y
0
(s)) = z
0
(s), i. e., we seek a surface o dened by
z = u(x, y) which contains the curve .
y
z
x
C
Figure 2.4: Cauchy initial value problem

Denition. The curve is said to be noncharacteristic if
x
0
(s)a
2
(x
0
(s), y
0
(s)) y
0
(s)a
1
(x
0
(s), y
0
(s)) ,= 0.
Theorem 2.1. Assume a
1
, a
2
, a
2
C
1
in their arguments, the initial data
x
0
, y
0
, z
0
C
1
[s
1
, s
2
] and is noncharacteristic.
Then there is a neighbourhood of ( such that there exists exactly one
solution u of the Cauchy initial value problem.
Proof. (i) Existence. Consider the following initial value problem for the
system of characteristic equations to ():
x
(t) = a
1
(x, y, z)
y
(t) = a
2
(x, y, z)
z
(t) = a
3
(x, y, z)
with the initial conditions
x(s, 0) = x
0
(s)
y(s, 0) = y
0
(s)
z(s, 0) = z
0
(s).
Let x = x(s, t), y = y(s, t), z = z(s, t) be the solution, s
1
s s
2
, [t[ <
for an > 0. We will show that this set of strings sticked onto the curve
, see Figure 2.4, denes a surface. To show this, we consider the inverse
functions s = s(x, y), t = t(x, y) of x = x(s, t), y = y(s, t) and show that
z(s(x, y), t(x, y)) is a solution of the initial problem of Cauchy. The inverse
functions s and t exist in a neighbourhood of t = 0 since
det
(x, y)
(s, t)
t=0
=
x
s
x
t
y
s
y
t
t=0
= x
0
(s)a
2
y
0
(s)a
1
,= 0,
and the initial curve is noncharacteristic by assumption.
Set
u(x, y) := z(s(x, y), t(x, y)),
then u satises the initial condition since
u(x, y)[
t=0
= z(s, 0) = z
0
(s).
The following calculation shows that u is also a solution of the dierential
equation ().
a
1
u
x
+a
2
u
y
= a
1
(z
s
s
x
+z
t
t
x
) +a
2
(z
s
s
y
+z
t
t
y
)
= z
s
(a
1
s
x
+a
2
s
y
) +z
t
(a
1
t
x
+a
2
t
y
)
= z
s
(s
x
x
t
+s
y
y
t
) +z
t
(t
x
x
t
+t
y
y
t
)
= a
3
since 0 = s
t
= s
x
x
t
+s
y
y
t
and 1 = t
t
= t
x
x
t
+t
y
y
t
.
(ii) Uniqueness. Suppose that v(x, y) is a second solution. Consider a point
(x
, y
) in a neighbourhood of the curve (x

0
(s), y(s)), s
1
s s
2
+ ,
> 0 small. The inverse parameters are s
= s(x
, y
), t
= t(x
, y
), see
Figure 2.5.
x
y
(x (s),y (s))
0 0
(x,y)
Figure 2.5: Uniqueness proof
Let
/ : x(t) := x(s
, t), y(t) := y(s
, t), z(t) := z(s
, t)
be the solution of the above initial value problem for the characteristic dif-
ferential equations with the initial data
x(s
, 0) = x
0
(s
), y(s
, 0) = y
0
(s
), z(s
, 0) = z
0
(s
).
According to its construction this curve is on the surface o dened by u =
u(x, y) and u(x
, y
) = z(s
, t
). Set
(t) := v(x(t), y(t)) z(t),
then
(t) = v
x
x
+v
y
y
= x
x
a
1
+v
y
a
2
a
3
= 0
and
(0) = v(x(s
, 0), y(s
, 0)) z(s
, 0) = 0
since v is a solution of the dierential equation and satises the initial con-
dition by assumption. Thus, (t) 0, i. e.,
v(x(s
, t), y(s
, t)) z(s
, t) = 0.
Set t = t
, then
v(x
, y
) z(s
, t
) = 0,
which shows that v(x
, y
) = u(x
, y
) because of z(s
, t
) = u(x
, y
). 2
Remark. In general, there is no uniqueness if the initial curve is a
characteristic curve, see an exercise and Figure 2.6 which illustrates this
case.
y
z
x
u
v
S
S
Figure 2.6: Multiple solutions
Examples
1. Consider the Cauchy initial value problem
u
x
+u
y
= 0
with the initial data
x
0
(s) = s, y
0
(s) = 1, z
0
(s) is a given C
1
-function.
These initial data are noncharacteristic since y
0
a
1
x
0
a
2
= 1. The solution
of the associated system of characteristic equations
x
(t) = 1, y
(t) = 1, u
(t) = 0
with the initial conditions
x(s, 0) = x
0
(s), y(s, 0) = y
0
(s), z(s, 0) = z
0
(s)
is given by
x = t +x
0
(s), y = t +y
0
(s), z = z
0
(s),
i. e.,
x = t +s, y = t + 1, z = z
0
(s).
It follows s = x y +1, t = y 1 and that u = z
0
(x y +1) is the solution
of the Cauchy initial value problem.
2. A problem from kinetics in chemistry. Consider for x 0, y 0 the
problem
u
x
+u
y
=
_
k
0
e
k
1
x
+k
2
_
(1 u)
with initial data
u(x, 0) = 0, x > 0, and u(0, y) = u
0
(y), y > 0.
Here the constants k
j
are positive, these constants dene the velocity of the
reactions in consideration, and the function u
0
(y) is given. The variable x
is the time and y is the hight of a tube, for example, in which the chemical
reaction takes place, and u is the concentration of the chemical substance.
In contrast to our previous assumptions, the initial data are not in C
1
.
The projection (
1
(
2
of the initial curve onto the (x, y)-plane has a corner
at the origin, see Figure 2.7.
x
y
x=y
2
1
C
C
1
2
Figure 2.7: Domains to the chemical kinetics example
The associated system of characteristic equations is
x
(t) = 1, y
(t) = 1, z
(t) =
_
k
0
e
k
1
x
+k
2
_
(1 z).
It follows x = t + c
1
, y = t + c
2
with constants c
j
. Thus the projection
of the characteristic curves on the (x, y)-plane are straight lines parallel to
y = x. We will solve the initial value problems in the domains
1
and
2
,
see Figure 2.7, separately.
(i)The initial value problem in
1
. The initial data are
x
0
(s) = s, y
0
(s) = 0, z
0
(0) = 0, s 0.
It follows
x = x(s, t) = t +s, y = y(s, t) = t.
Thus
z
(t) = (k
0
e
k
1
(t+s)
+k
2
)(1 z), z(0) = 0.
The solution of this initial value problem is given by
z(s, t) = 1 exp
_
k
0
k
1
e
k
1
(s+t)
k
2
t
k
0
k
1
e
k
1
s
_
.
Consequently
u
1
(x, y) = 1 exp
_
k
0
k
1
e
k
1
x
k
2
y k
0
k
1
e
k
1
(xy)
_
is the solution of the Cauchy initial value problem in
1
. If time x tends to
, we get the limit
lim
x
u
1
(x, y) = 1 e
k
2
y
.
(ii) The initial value problem in
2
. The initial data are here
x
0
(s) = 0, y
0
(s) = s, z
0
(0) = u
0
(s), s 0.
It follows
x = x(s, t) = t, y = y(s, t) = t +s.
Thus
z
(t) = (k
0
e
k
1
t
+k
2
)(1 z), z(0) = 0.
The solution of this initial value problem is given by
z(s, t) = 1 (1 u
0
(s)) exp
_
k
0
k
1
e
k
1
t
k
2
t
k
0
k
1
_
.
Consequently
u
2
(x, y) = 1 (1 u
0
(y x)) exp
_
k
0
k
1
e
k
1
x
k
2
x
k
0
k
1
_
is the solution in
2
.
If x = y, then
u
1
(x, y) = 1 exp
_
k
0
k
1
e
k
1
x
k
2
x
k
0
k
1
_
u
2
(x, y) = 1 (1 u
0
(0)) exp
_
k
0
k
1
e
k
1
x
k
2
x
k
0
k
1
_
.
If u
0
(0) > 0, then u
1
< u
2
if x = y, i. e., there is a jump of the concentration
of the substrate along its burning front dened by x = y.
Remark. Such a problem with discontinuous initial data is called Riemann
problem. See an exercise for another Riemann problem.
The case that a solution of the equation is known
Here we will see that we get immediately a solution of the Cauchy initial
value problem if a solution of the homogeneous linear equation
a
1
(x, y)u
x
+a
2
(x, y)u
y
= 0
is known.
Let
x
0
(s), y
0
(s), z
0
(s), s
1
< s < s
2
be the initial data and let u = (x, y) be a solution of the dierential equa-
tion. We assume that
x
(x
0
(s), y
0
(s))x
0
(s) +
y
(x
0
(s), y
0
(s))y
0
(s) ,= 0
is satised. Set g(s) = (x
0
(s), y
0
(s)) and let s = h(g) be the inverse
function.
The solution of the Cauchy initial problem is given by u
0
(h((x, y))).
This follows since in the problem considered a composition of a solution is
a solution again, see an exercise, and since
u
0
(h((x
0
(s), y
0
(s))) = u
0
(h(g)) = u
0
(s).
Example: Consider equation
u
x
+u
y
= 0
with initial data
x
0
(s) = s, y
0
(s) = 1, u
0
(s) is a given function.
A solution of the dierential equation is (x, y) = x y. Thus
((x
0
(s), y
0
(s)) = s 1
and
u
0
( + 1) = u
0
(x y + 1)
is the solution of the problem.
2.3 Nonlinear equations in two variables
Here we consider equation
F(x, y, z, p, q) = 0, (2.6)
where z = u(x, y), p = u
x
(x, y), q = u
y
(x, y) and F C
2
is given such that
F
2
p
+F
2
q
,= 0.
In contrast to the quasilinear case, this general nonlinear equation is
more complicated. Together with (2.6) we will consider the following system
of ordinary equations which follow from considerations below as necessary
conditions, in particular from the assumption that there is a solution of
(2.6).
x
(t) = F
p
(2.7)
y
(t) = F
q
(2.8)
z
(t) = pF
p
+qF
q
(2.9)
p
(t) = F
x
F
u
p (2.10)
q
(t) = F
y
F
u
q. (2.11)
2.3. NONLINEAR EQUATIONS IN TWO VARIABLES 41
Denition. Equations (2.7)(2.11) are said to be characteristic equations
of equation (2.6) and a solution
(x(t), y(t), z(t), p(t), q(t))
of the characteristic equations is called characteristic strip or Monge curve.
Figure 2.8: Gaspard Monge (Pantheon, Paris)
We will see, as in the quasilinear case, that the strips dened by the char-
acteristic equations build the solution surface of the Cauchy initial value
problem.
Let z = u(x, y) be a solution of the general nonlinear dierential equa-
tion (2.6).
Let (x
0
, y
0
, z
0
) be xed, then equation (2.6) denes a set of planes given
by (x
0
, y
0
, z
0
, p, q), i. e., planes given by z = v(x, y) which contain the point
(x
0
, y
0
, z
0
) and for which v
x
= p, v
y
= q at (x
0
, y
0
). In the case of quasilinear
equations these set of planes is a bundle of planes which all contain a xed
straight line, see Section 2.1. In the general case of this section the situation
is more complicated.
Consider the example
p
2
+q
2
= f(x, y, z), (2.12)
where f is a given positive function. Let E be a plane dened by z = v(x, y)
and which contains (x
0
, y
0
, z
0
). Then the normal on the plane E directed
downward is
N =
1
_
1 +[v[
2
(p, q, 1),
where p = v
x
(x
0
, y
0
), q = v
y
(x
0
, y
0
). It follows from (2.12) that the normal
N makes a constant angle with the z-axis, and the z-coordinate of N is
constant, see Figure 2.9.
y
z
x
N
()
()
Figure 2.9: Monge cone in an example
Thus the endpoints of the normals xed at (x
0
, y
0
, z
0
) dene a circle
parallel to the (x, y)-plane, i. e., there is a cone which is the envelope of all
these planes.
We assume that the general equation (2.6) denes such a Monge cone at
each point in R
3
. Then we seek a surface S which touches at each point its
Monge cone, see Figure 2.10.
More precisely, we assume there exists, as in the above example, a one
parameter C
1
-family
p() = p(; x, y, z), q() = q(; x, y, z)
of solutions of (2.6). These (p(), q()) dene a family () of planes.
y
z
x
Figure 2.10: Monge cones
Let
x() = (x(), y(), z())
be a curve on the surface S which touches at each point its Monge cone,
see Figure 2.11. Thus we assume that at each point of the surface o the
associated tangent plane coincides with a plane from the family () at
this point. Consider the tangential plane T
x
0
of the surface S at x
0
=
(x(
0
), y(
0
), z(
0
)). The straight line
l() = x
0
+x
(
0
), < < ,
is an apothem (in German: Mantellinie) of the cone by assumption and is
contained in the tangential plane T
x
0
as the tangent of a curve on the surface
S. It is dened through
x
(
0
) = l
(). (2.13)
The straight line l() satises
l
3
() z
0
= (l
1
() x
0
)p(
0
) + (l
2
() y
0
)q(
0
),
since it is contained in the tangential plane T
x
0
dened by the slope (p, q).
It follows
l
3
() = p(
0
)l
1
() +q(
0
)l
2
().
y
z
x
S
T
x
0
Figure 2.11: Monge cones along a curve on the surface
Together with (2.13) we obtain
z
() = p(
0
)x
() +q(
0
)y
(). (2.14)
The above straight line l is the limit of the intersection line of two neigh-
bouring planes which envelopes the Monge cone:
z z
0
= (x x
0
)p(
0
) + (y y
0
)q(
0
)
z z
0
= (x x
0
)p(
0
+h) + (y y
0
)q(
0
+h).
On the intersection one has
(x x
0
)p() + (y y
0
)q(
0
) = (x x
0
)p(
0
+h) + (y y
0
)q(
0
+h).
Let h 0, it follows
(x x
0
)p
(
0
) + (y y
0
)q
(
0
) = 0.
Since x = l
1
(), y = l
2
() in this limit position, we have
p
(
0
)l
1
() +q
(
0
)l
2
() = 0,
and it follows from (2.13) that
p
(
0
)x
() +q
(
0
)y
() = 0. (2.15)
From the dierential equation F(x
0
, y
0
, z
0
, p(), q()) = 0 we see that
F
p
p
() +F
q
q
() = 0. (2.16)
Assume x
(
0
) ,= 0 and F
p
,= 0, then we obtain from (2.15), (2.16)
y
(
0
)
x
(
0
)
=
F
q
F
p
,
and from (2.14) (2.16) that
z
(
0
)
x
(
0
)
= p +q
F
q
F
p
.
It follows, since
0
was an arbitrary xed parameter,
x
() = (x
(), y
(), z
())
=
_
x
(), x
()
F
q
F
p
, x
()
_
p +q
F
q
F
p
__
=
x
()
F
p
(F
p
, F
q
, pF
p
+qF
q
),
i. e., the tangential vector x
() is proportional to (F
p
, F
q
, pF
p
+qF
q
). Set
a() =
x
()
F
p
,
where F = F(x(), y(), z(), p(()), q(())). Introducing the new pa-
rameter t by the inverse of = (t), where
t() =
_

0
a(s) ds,
we obtain the characteristic equations (2.7)(2.9). Here we denote x((t))
by x(t) again. From the dierential equation (2.6) and from (2.7)(2.9)
we get equations (2.10) and (2.11). Assume the surface z = u(x, y) under
consideration is in C
2
, then
F
x
+F
z
p +F
p
p
x
+F
q
p
y
= 0, (q
x
= p
y
)
F
x
+F
z
p +x
(t)p
x
+y
(t)p
y
= 0
F
x
+F
z
p +p
(t) = 0
since p = p(x, y) = p(x(t), y(t)) on the curve x(t). Thus equation (2.10) of
the characteristic system is shown. Dierentiating the dierential equation
(2.6) with respect to y, we get nally equation (2.11).
Remark. In the previous quasilinear case
F(x, y, z, p, q) = a
1
(x, y, z)p +a
2
(x, y, z)q a
3
(x, y, z)
the rst three characteristic equations are the same:
x
(t) = a
1
(x, y, z), y
(t) = a
2
(x, y, z), z
(t) = a
3
(x, y, z).
The point is that the right hand sides are independent on p or q. It follows
from Theorem 2.1 that there exists a solution of the Cauchy initial value
problem provided the initial data are noncharacteristic. That is, we do not
need the other remaining two characteristic equations.
The other two equations (2.10) and (2.11) are satised in this quasilin-
ear case automatically if there is a solution of the equation, see the above
derivation of these equations.
The geometric meaning of the rst three characteristic dierential equa-
tions (2.7)(2.11) is the following one. Each point of the curve
/ : (x(t), y(t), z(t)) corresponds a tangential plane with the normal direc-
tion (p, q, 1) such that
z
(t) = p(t)x
(t) +q(t)y
(t).
This equation is called strip condition. On the other hand, let z = u(x, y)
denes a surface, then z(t) := u(x(t), y(t)) satises the strip condition, where
p = u
x
and q = u
y
, that is, the scales dened by the normals t together.
Proposition 2.3. F(x, y, z, p, q) is an integral, i. e., it is constant along
each characteristic curve.
Proof.
d
dt
F(x(t), y(t), z(t), p(t), q(t)) = F
x
x
+F
y
y
+F
z
z
+F
p
p
+F
q
q
= F
x
F
p
+F
y
F
q
+pF
z
F
p
+qF
z
F
q
F
p
f
x
F
p
F
z
p F
q
F
y
F
q
F
z
q
= 0.
2
Corollary. Assume F(x
0
, y
0
, z
0
, p
0
, q
0
) = 0, then F = 0 along characteristic
curves with the initial data (x
0
, y
0
, z
0
, p
0
, q
0
).
Proposition 2.4. Let z = u(x, y), u C
2
, be a solution of the nonlinear
equation (2.6). Set
z
0
= u(x
0
, y
0
, ) p
0
= u
x
(x
0
, y
0
), q
0
= u
y
(x
0
, y
0
).
Then the associated characteristic strip is in the surface o, dened by z =
u(x, y). Thus
z(t) = u(x(t), y(t))
p(t) = u
x
(x(t), y(t))
q(t) = u
y
(x(t), y(t)),
where (x(t), y(t), z(t), p(t), q(t)) is the solution of the characteristic system
(2.7)(2.11) with initial data (x
0
, y
0
, z
0
, p
0
, q
0
)
Proof. Consider the initial value problem
x
(t) = F
p
(x, y, u(x, y), u
x
(x, y), u
y
(x, y))
y
(t) = F
q
(x, y, u(x, y), u
x
(x, y), u
y
(x, y))
with the initial data x(0) = x
0
, y(0) = y
0
. We will show that
(x(t), y(t), u(x(t), y(t)), u
x
(x(t), y(t)), u
y
(x(t), y(t)))
is a solution of the characteristic system. We recall that the solution exists
and is uniquely determined.
Set z(t) = u(x(t), y(t)), then (x(t), y(t), z(t)) o, and
z
(t) = u
x
x
(t) +u
y
y
(t) = u
x
F
p
+u
y
F
q
.
Set p(t) = u
x
(x(t), y(t)), q(t) = u
y
(x(t), y(t)), then
p
(t) = u
xx
F
p
+u
xy
F
q
q
(t) = u
yx
F
p
+u
yy
F
q
.
Finally, from the dierential equation F(x, y, u(x, y), u
x
(x, y), u
y
(x, y)) = 0
it follows
p
(t) = F
x
F
u
p
q
(t) = F
y
F
u
q.
2
2.3.1 Initial value problem of Cauchy
Let
x = x
0
(s), y = y
0
(s), z = z
0
(s), p = p
0
(s), q = q
0
(s), s
1
< s < s
2
, (2.17)
be a given initial strip such that the strip condition
z
0
(s) = p
0
(s)x
0
(s) +q
0
(s)y
0
(s) (2.18)
is satised. Moreover, we assume that the initial strip satises the nonlinear
equation, that is,
F(x
0
(s), y
0
(s), z
0
(s), p
0
(s), q
0
(s)) = 0. (2.19)
Initial value problem of Cauchy: Find a C
2
-solution z = u(x, y) of
F(x, y, z, p, q) = 0 such that the surface o dened by z = u(x, y) contains
the above initial strip.
Similar to the quasilinear case we will show that the set of strips de-
ned by the characteristic system which are sticked at the initial strip, see
Figure 2.12, t together and dene the surface for which we are looking at.
Denition. A strip (x(), y(), z(), p(), q()),
1
< <
2
, is said to be
noncharacteristic if
x
()F
q
(x(), y(), z(), p(), q())y
()F
p
(x(), y(), z(), p(), q()) ,= 0.
Theorem 2.2. For a given noncharacteristic initial strip (2.17), x
0
, y
0
, z
0

C
2
and p
0
, q
0
C
1
which satises the strip condition (2.18) and the dif-
ferential equation (2.19) there exists exactly one solution z = u(x, y) of
the Cauchy initial value problem in a neighbourhood of the initial curve
(x
0
(s), y
0
(s), z
0
(s)), i. e., z = u(x, y) is the solution of the dierential equa-
tion (2.6) and u(x
0
(s), y
0
(s)) = z
0
(s), u
x
(x
0
(s), y
0
(s)) = p
0
(s), u
y
(x
0
(s), y
0
(s)) =
q
0
(s).
Proof. Consider the system (2.7)(2.11) with initial data
x(s, 0) = x
0
(s), y(s, 0) = y
0
(s), z(s, 0) = z
0
(s), p(s, 0) = p
0
(s), q(s, 0) = q
0
(s).
We will show that the surface dened by x = x(s, t), y(s, t) is the surface
dened by z = u(x, y), where u is the solution of the Cauchy initial value
y
z
x
t=0
t>0
Figure 2.12: Construction of the solution
problem. It turns out that u(x, y) = z(s(x, y), t(x, y)), where s = s(x, y),
t = t(x, y) is the inverse of x = x(s, t), y = y(s, t) in a neigbourhood of t = 0.
This inverse exists since the initial strip is noncharacteristic by assumption:
det
(x, y)
(s, t)
t=0
= x
0
F
q
y
0
F
q
,= 0.
Set
P(x, y) = p(s(x, y), t(x, y)), Q(x, y) = q(s(x, y), t(x, y)).
From Proposition 2.3 and Proposition 2.4 it follows F(x, y, u, P, Q) = 0. We
will show that P(x, y) = u
x
(x, y) and Q(x, y) = u
y
(x, y). To see this, we
consider the function
h(s, t) = z
s
px
s
qy
s
.
One has
h(s, 0) = z
0
(s) p
0
(s)x
0
(s) q
0
(s)y
0
(s) = 0
since the initial strip satises the strip condition by assumption. In the
following we will nd that for xed s the function h satises a linear ho-
mogeneous ordininary dierential equation of rst order. Consequently,
h(s, t) = 0 in a neighbourhood of t = 0. Thus the strip condition is also sat-
ised along strips transversally to the characteristic strips, see Figure 2.18.
Thaen the set of scales t together and dene a surface like the scales of
a sh.
From the denition of h(s, t) and the characteristic equations we get
h
t
(s, t) = z
st
p
t
x
s
q
t
y
s
px
st
qy
st
=

s
(z
t
px
t
qy
t
) +p
s
x
t
+q
s
y
t
q
t
y
s
p
t
x
s
= (px
s
+qy
s
)F
z
+F
x
x
s
+F
y
z
s
+F
p
p
s
+F
q
q
s
.
Since F(x(s, t), y(s, t), z(s, t), p(s, t), q(s, t)) = 0, it follows after dierentia-
tion of this equation with respect to s the dierential equation
h
t
= F
z
h.
Hence h(s, t) 0, since h(s, 0) = 0.
Thus we have
z
s
= px
s
+qy
s
z
t
= px
t
+qy
t
z
s
= u
x
x
s
+u
y
y
s
z
t
= u
x
y
t
+u
y
y
t
.
The rst equation was shown above, the second is a characteristic equation
and the last two follow from z(s, t) = u(x(s, t), y(s, t)). This system implies
(P u
x
)x
s
+ (Qu
y
)y
s
= 0
(P u
x
)x
t
+ (Qu
y
)y
t
= 0.
It follows P = u
x
and Q = u
y
.
The initial conditions
u(x(s, 0), y(s, 0)) = z
0
(s)
u
x
(x(s, 0), y(s, 0)) = p
0
(s)
u
y
(x(s, 0), y(s, 0)) = q
0
(s)
are satised since
u(x(s, t), y(s, t)) = z(s(x, y), t(x, y)) = z(s, t)
u
x
(x(s, t), y(s, t)) = p(s(x, y), t(x, y)) = p(s, t)
u
y
(x(s, t), y(s, t)) = q(s(x, y), t(x, y)) = q(s, t).
2.4. NONLINEAR EQUATIONS IN R
N
51
The uniqueness follows as in the proof of Theorem 2.1. 2
Example. A dierential equation which occurs in the geometrical optic is
u
2
x
+u
2
y
= f(x, y),
where the positive function f(x, y) is the index of refraction. The level sets
dened by u(x, y) = const. are called wave fronts. The characteristic curves
(x(t), y(t)) are the rays of light. If n is a constant, then the rays of light are
straight lines. In R
3
the equation is
u
2
x
+u
2
y
+u
2
z
= f(x, y, z).
Thus we have to extend the previous theory from R
2
to R
n
, n 3.
2.4 Nonlinear equations in R
n
Here we consider the nonlinear dierential equation
F(x, z, p) = 0, (2.20)
where
x = (x
1
, . . . , x
n
), z = u(x) : R
n
R, p = u.
The following system of 2n+1 ordinary dierential equations is called char-
acteristic system.
x
(t) =
p
F
z
(t) = p
p
F
p
(t) =
x
F F
z
p.
Let
x
0
(s) = (x
01
(s), . . . , x
0n
(s)), s = (s
1
, . . . , s
n1
),
be a given regular (n-1)-dimensional C
2
-hypersurface in R
n
, i. e., we assume
rank
x
0
(s)
s
= n 1.
Here s D is a parameter from an (n 1)-dimensional parameter domain
D.
For example, x = x
0
(s) denes in the three dimensional case a regular
surface in R
3
.
Assume
z
0
(s) : D R, p
0
(s) = (p
01
(s), . . . , p
0n
(s))
are given suciently regular functions.
The (2n + 1)-vector
(x
0
(s), z
0
(s), p
0
(s))
is called initial strip manifold and the condition
z
0
s
l
=
n1
i=1
p
0i
(s)
x
0i
s
l
,
l = 1, . . . , n 1, strip condition.
The initial strip manifold is said to be noncharacteristic if
det
_
_
_
_
_
F
p
1
F
p
2
F
p
n
x
01
s
1
x
02
s
1

x
0n
s
1
. . . . . . . . . . . . . . . . . . . . . . . . .
x
01
s
n1
x
02
s
n1

x
0n
s
n1
_
_
_
_
_
,= 0,
where the argument of F
p
j
is the initial strip manifold.
Initial value problem of Cauchy. Seek a solution z = u(x) of the
dierential equation (2.20) such that the initial manifold is a subset of
(x, u(x), u(x)) : x .
As in the two dimensional case we have under additional regularity as-
sumptions
Theorem 2.3. Suppose the initial strip manifold is not characteristic and
satises dierential equation (2.20), that is, F(x
0
(s), z
0
(s), p
0
(s)) = 0. Then
there is a neighbourhood of the initial manifold (x
0
(s), z
0
(s)) such that there
exists a unique solution of the Cauchy initial value problem.
Sketch of proof. Let
x = x(s, t), z = z(s, t), p = p(s, t)
be the solution of the characteristic system and let
s = s(x), t = t(x)
2.5. HAMILTON-JACOBI THEORY 53
be the inverse of x = x(s, t) which exists in a neighbourhood of t = 0. Then,
it turns out that
z = u(x) := z(s
1
(x
1
, . . . , x
n
), . . . , s
n1
(x
1
, . . . , x
n
), t(x
1
, . . . , x
n
))
is the solution of the problem.
2.5 Hamilton-Jacobi theory
The nonlinear equation (2.20) of previous section in one more dimension is
F(x
1
, . . . , x
n
, x
n+1
, z, p
1
, . . . , p
n
, p
n+1
) = 0.
The content of the Hamilton
1
-Jacobi
2
theory is the theory of the special
case
F p
n+1
+H(x
1
, . . . , x
n
, x
n+1
, p
1
, . . . , p
n
) = 0, (2.21)
i. e., the equation is linear in p
n+1
and does not depend on z explicitly.
Remark. Formally, one can write equation (2.20)
F(x
1
, . . . , x
n
, u, u
x
1
, . . . , u
x
n
) = 0
as an equation of type (2.21). Set x
n+1
= u and seek u implicitely from
(x
1
, . . . , x
n
, x
n+1
) = const.,
where is a function which is dened by a dierential equation.
Assume
x
n+1
,= 0, then
0 = F(x
1
, . . . , x
n
, u, u
x
1
, . . . , u
x
n
)
= F(x
1
, . . . , x
n
, x
n+1
,

x
1
x
n+1
, . . . ,

x
n
x
n+1
)
= : G(x
1
, . . . , x
n+1
,
1
, . . . ,
x
n+1
).
Suppose that G
x
n+1
,= 0, then
x
n+1
= H(x
1
, . . . , x
n
, x
n+1
,
x
1
, . . . ,
x
n+1
).
1
Hamilton, William Rowan, 18051865
2
Jacobi, Carl Gustav, 18051851
The associated characteristic equations to (2.21) are
x
n+1
() = F
p
n+1
= 1
x
k
() = F
p
k
= H
p
k
, k = 1, . . . , n
z
() =
n+1
l=1
p
l
F
p
l
=
n
l=1
p
l
H
p
l
+p
n+1
=
n
l=1
p
l
H
p
l
H
p
n+1
() = F
x
n+1
F
z
p
n+1
= F
x
n+1
p
k
() = F
x
k
F
z
p
k
= F
x
k
, k = 1, . . . , n.
Set t := x
n+1
, then we can write partial dierential equation (2.21) as
u
t
+H(x, t,
x
u) = 0 (2.22)
and 2n of the characteristic equations are
x
(t) =
p
H(x, t, p) (2.23)
p
(t) =
x
H(x, t, p). (2.24)
Here is
x = (x
1
, . . . , x
n
), p = (p
1
, . . . , p
n
).
Let x(t), p(t) be a solution of (2.23) and (2.24), then it follows p
n+1
(t) and
z
(t) from the characteristic equations

p
n+1
(t) = H
t
z
(t) = p
p
H H.
Denition. The function H(x, t, p) is called Hamilton function, equa-
tion (2.21) Hamilton-Jacobi equation and the system (2.23), (2.24) canonical
system to H.
There is an interesting interplay between the Hamilton-Jacobi equation
and the canonical system. According to the previous theory we can con-
struct a solution of the Hamilton-Jacobi equation by using solutions of the
canonical system. On the other hand, one obtains from solutions of the
Hamilton-Jacobi equation also solutions of the canonical system of ordinary
dierential equations.
Denition. A solution (a; x, t) of the Hamilton-Jacobi equation, where
a = (a
1
, . . . , a
n
) is an n-tuple of real parameters, is called a complete integral
of the Hamilton-Jacobi equation if
det(
x
i
a
l
)
n
i,l=1
,= 0.
Remark. If u is a solution of the Hamilton-Jacobi equation, then also
u +const.
Theorem 2.4 (Jacobi). Assume
u = (a; x, t) +c, c = const., C
2
in its arguments,
is a complete integral. Then one obtains by solving of
b
i
=
a
i
(a; x, t)
with respect to x
l
= x
l
(a, b, t), where b
i
i = 1, . . . , n are given real constants,
and then by setting
p
k
=
x
k
(a; x(a, b; t), t)
a 2n-parameter family of solutions of the canonical system.
Proof. Let
x
l
(a, b; t), l = 1, . . . , n,
be the solution of the above system. The solution exists since is a complete
integral by assumption. Set
p
k
(a, b; t) =
x
k
(a; x(a, b; t), t), k = 1, . . . , n.
We will show that x and p solves the canonical system. Dierentiating
a
i
=
b
i
with respect to t and the Hamilton-Jacobi equation
t
+H(x, t,
x
) = 0
with respect to a
i
, we obtain for i = 1, . . . , n
ta
i
+
n
k=1
x
k
a
i
x
k
t
= 0
ta
i
+
n
k=1
x
k
a
i
Hp
k
= 0.
Since is a complete integral it follows for k = 1, . . . , n
x
k
t
= H
p
k
.
Along a trajectory, i. e., where a, b are xed, it is
x
k
t
= x
k
(t). Thus
x
k
(t) = H
p
k
.
Now we dierentiate p
i
(a, b; t) with respect to t and
t
+ H(x, t,
x
) = 0
with respect to x
i
, and obtain
p
i
(t) =
x
i
t
+
n
k=1
x
i
x
k
x
k
(t)
0 =
x
i
t
+
n
k=1
x
i
x
k
H
p
k
+H
x
i
0 =
x
i
t
+
n
k=1
x
i
x
k
x
k
(t) +H
x
i
It follows nally that p
i
(t) = H
x
i
. 2
Example: Kepler problem
The motion of a mass point in a central eld takes place in a plane,
say the (x, y)-plane, see Figure 2.13, and satises the system of ordinary
dierential equations of second order
x
(t) = U
x
, y
(t) = U
y
,
where
U(x, y) =
k
2
_
x
2
+y
2
.
Here we assume that k
2
is a positive constant and that the mass point is
attracted of the origin. In the case that it is pushed one has to replace U
by U. See Landau and Lifschitz [12], Vol 1, for example, for the related
physics.
Set
p = x
, q = y
and
H =
1
2
(p
2
+q
2
) U(x, y),
x
y
(x(t),y(t))
(U ,U )
y
x
Figure 2.13: Motion in a central eld

then
x
(t) = H
p
, y
(t) = H
q
p
(t) = H
x
, q
(t) = H
y
.
The associated Hamilton-Jacobi equation is
t
+
1
2
(
2
x
+
2
y
) =
k
2
_
x
2
+y
2
.
which is in polar coordinates (r, )
t
+
1
2
(
2
r
+
1
r
2
) =
k
2
r
. (2.25)
Now we will seek a complete integral of (2.25) by making the ansatz
t
= = const.
= = const. (2.26)
and obtain from (2.25) that
=
_
r
r
0
2 +
2k
2
2
d +c(t, ).
From ansatz (2.26) it follows
c(t, ) = t .
Therefore we have a two parameter family of solutions
= (, ; , r, t)
of the Hamilton-Jacobi equation. This solution is a complete integral, see
an exercise. According to the theorem of Jacobi set
= t
0
,
=
0
.
Then
t t
0
=
_
r
r
0
d
_
2 +
2k
2
2
.
The inverse function r = r(t), r(0) = r
0
, is the r-coordinate depending on
time t, and

0
=
_
r
r
0
d
2
_
2 +
2k
2
2
.
Substitution =
1
yields

0
=
_
1/r
1/r
0
d
_
2 + 2k
2

2
2
= arcsin
_

2
k
2
1
r
1
_
1 +
2
2
k
4
_
+ arcsin
_

2
k
2
1
r
0
1
_
1 +
2
2
k
4
_
.
Set
1
=
0
+ arcsin
_

2
k
2
1
r
0
1
_
1 +
2
2
k
4
_
and
p =

2
k
2
,
2
=
_
1 +
2
2
k
4
,
then

1
= arcsin
_
p
r
1
2
_
.
It follows
r = r() =
p
1
2
sin(
1
)
,
which is the polar equation of conic sections. It denes an ellipse if 0 < 1,
a parabola if = 1 and a hyperbola if > 1, see Figure 2.14 for the case
of an ellipse, where the origin of the coordinate system is one of the focal
points of the ellipse.
For another application of the Jacobi theorem see Courant and Hilbert [4],
Vol. 2, pp. 94, where geodedics on an ellipsoid are studied.
2.6. EXERCISES 59
1
p
p
1+
1
2
2
p
Figure 2.14: The case of an ellipse
2.6 Exercises
1. Suppose u : R
2
R is a solution of
a(x, y)u
x
+b(x, y)u
y
= 0.
Show that for arbitrary H C
1
also H(u) is a solution.
2. Find a solution u , const. of
u
x
+u
y
= 0
such that
graph(u) := (x, y, z) R
3
: z = u(x, y), (x, y) R
2
contains the straight line (0, 0, 1) +s(1, 1, 0), s R.

3. Let (x, y) be a solution of
a
1
(x, y)u
x
+a
2
(x, y)u
y
= 0 .
Prove that level curves S
C
:= (x, y) : (x, y) = C = const. are
characteristic curves, provided that ,= 0 and (a
1
, a
2
) ,= (0, 0).
4. Prove Proposition 2.2.
5. Find two dierent solutions of the initial value problem
u
x
+u
y
= 1,
where the initial data are x
0
(s) = s, y
0
(s) = s, z
0
(s) = s.
Hint: (x
0
, y
0
) is a characteristic curve.
6. Solve the initial value problem
xu
x
+yu
y
= u
with initial data x
0
(s) = s, y
0
(s) = 1, z
0
(s), where z
0
is given.
xu
x
+yu
y
= xu
2
,
x
0
(s) = s, y
0
(s) = 1, z
0
(s) = e
s
.
uu
x
+u
y
= 1,
x
0
(s) = s, y
0
(s) = s, z
0
(s) = s/2 if 0 < s < 1.
uu
x
+uu
y
= 2,
x
0
(s) = s, y
0
(s) = 1, z
0
(s) = 1 +s if 0 < s < 1.
10. Solve the initial value problem u
2
x
+u
2
y
= 1 +x with given initial data
x
0
(s) = 0, y
0
(s) = s, u
0
(s) = 1, p
0
(s) = 1, q
0
(s) = 0, < s < .
11. Find the solution (x, y) of
(x y)u
x
+ 2yu
y
= 3x
such that the surface dened by z = (x, y) contains the curve
C : x
0
(s) = s, y
0
(s) = 1, z
0
(s) = 0, s R.
2.6. EXERCISES 61
12. Solve the following initial problem of chemical kinetics.
u
x
+u
y
=
_
k
0
e
k
1
x
+k
2
_
(1 u)
2
, x > 0, y > 0
with the initial data u(x, 0) = 0, u(0, y) = u
0
(y), where u
0
, 0 < u
0
< 1,
is given.
13. Solve the Riemann problem
u
x
1
+u
x
2
= 0
u(x
1
, 0) = g(x
1
)
in
1
= (x
1
, x
2
) R
2
: x
1
> x
2
and in
2
= (x
1
, x
2
) R
2
: x
1
<
x
2
, where
g(x
1
) =
_
u
l
: x
1
< 0
u
r
: x
1
> 0
with constants u
l
,= u
r
.
14. Determine the opening angle of the Monge cone, i. e., the angle be-
tween the axis and the apothem (in German: Mantellinie) of the cone,
for equation
u
2
x
+u
2
y
= f(x, y, u),
where f > 0.
u
2
x
+u
2
y
= 1,
where x
0
() = a cos , y
0
() = a sin , z
0
() = 1, p
0
() = cos ,
q
0
() = sin if 0 < 2, a = const. > 0.
16. Show that the integral (, ; , r, t), see the Kepler problem, is a
complete integral.
17. a) Show that S =

x +
1 y + , , R, 0 < < 1, is a
complete integral of S
x

_
1 S
2
y
= 0.
b) Find the envelope of this family of solutions.
18. Determine the length of the half axis of the ellipse
r =
p
1
2
sin(
0
)
, 0 < 1.
19. Find the Hamilton function H(x, p) of the Hamilton-Jacobi-Bellman
dierential equation if h = 0 and f = Ax + B, where A, B are
constant and real matrices, A : R
m
R
n
, B is an orthogonal real
n n-Matrix and p R
n
is given. The set of admissible controls is
given by
U = R
n
:
n
i=1
2
i
1 .
Remark. The Hamilton-Jacobi-Bellman equation is formally the Hamilton-
Jacobi equation u
t
+ H(x, u) = 0, where the Hamilton function is
dened by
H(x, p) := min
U
(f(x, ) p +h(x, )) ,
f(x, ) and h(x, ) are given. See for example, Evans [5], Chapter 10.
Chapter 3
Classication
Dierent types of problems in physics, for example, correspond dierent
types of partial dierential equations. The methods how to solve these
equations dier from type to type.
The classication of dierential equations follows from one single ques-
tion: Can we calculate formally the solution if suciently many initial data
are given? Consider the initial problem for an ordinary dierential equa-
tion y
(x) = f(x, y(x)), y(x

0
) = y
0
. Then one can determine formally the
solution, provided the function f(x, y) is suciently regular. The solution
of the initial value problem is formally given by a power series. This formal
solution is a solution of the problem if f(x, y) is real analytic according to
a theorem of Cauchy. In the case of partial dierential equations the re-
lated theorem is the Theorem of Cauchy-Kowalevskaya. Even in the case
of ordinary dierential equations the situation is more complicated if y
is
implicitly dened, i. e., the dierential equation is F(x, y(x), y
(x)) = 0 for
a given function F.
3.1 Linear equations of second order
The general nonlinear partial dierential equation of second order is
F(x, u, Du, D
2
u) = 0,
where x R
n
, u : R
n
R, Du u and D
2
u stands for all second
derivatives. The function F is given and suciently regular with respect to
its 2n + 1 +n
2
arguments.
63
64 CHAPTER 3. CLASSIFICATION
In this section we consider the case
n
i,k=1
a
ik
(x)u
x
i
x
k
+f(x, u, u) = 0. (3.1)
The equation is linear if
f =
n
i=1
b
i
(x)u
x
i
+c(x)u +d(x).
Concerning the classication the main part
n
i,k=1
a
ik
(x)u
x
i
x
k
plays the essential role. Suppose u C
2
, then we can assume, without
restriction of generality, that a
ik
= a
ki
, since
n
i,k=1
a
ik
u
x
i
x
k
=
n
i,k=1
(a
ik
)
u
x
i
x
k
,
where
(a
ik
)
=
1
2
(a
ik
+a
ki
).
Consider a hypersurface o in R
n
dened implicitly by (x) = 0, ,= 0,
see Figure 3.1
Assume u and u are given on o.
Problem: Can we calculate all other derivatives of u on o by using dier-
ential equation (3.1) and the given data?
We will nd an answer if we map o onto a hyperplane o
0
by a mapping
n
= (x
1
, . . . , x
n
)
i
=
i
(x
1
, . . . , x
n
), i = 1, . . . , n 1,
for functions
i
such that
det
(
1
, . . . ,
n
)
(x
1
, . . . , x
n
)
,= 0
in R
n
. It is assumed that and
i
are suciently regular. Such a
mapping = (x) exists, see an exercise.
3.1. LINEAR EQUATIONS OF SECOND ORDER 65
x
S
x
x
3
1
2
Figure 3.1: Initial manifold o
The above transform maps o onto a subset of the hyperplane dened by
n
= 0, see Figure 3.2.
We will write the dierential equation in these new coordinates. Here we
use Einsteins convention, i. e., we add terms with repeating indices. Since
u(x) = u(x()) =: v() = v((x)),
where x = (x
1
, . . . , x
n
) and = (
1
, . . . ,
n
), we get
u
x
j
= v
i
x
j
, (3.2)
u
x
j
x
k
= v
i
x
j
l
x
k
+v
i
x
j
x
k
.
Thus, dierential equation (3.1) in the new coordinates is given by
a
jk
(x)
i
x
j
l
x
k
v
l
+ terms known on o
0
= 0.
Since v
k
(
1
, . . . ,
n1
, 0), k = 1, . . . , n, are known, see (3.2), it follows that
v
l
, l = 1, . . . , n1, are known on o
0
. Thus we know all second derivatives
v
j
on o
0
with the only exception of v
n
.
3
1
2
S
0
Figure 3.2: Transformed at manifold o
0
We recall that, provided v is suciently regular,
v
l
(
1
, . . . ,
n1
, 0)
is the limit of
v
k
(
1
, . . . ,
l
+h,
l+1
, . . . ,
n1
, 0) v
k
(
1
, . . . ,
l
,
l+1
, . . . ,
n1
, 0)
h
as h 0.
Thus the dierential equation can be written as
n
j,k=1
a
jk
(x)
n
x
j
n
x
k
v
n
= terms known on o
0
.
It follows that we can calculate v
n
if
n
i,j=1
a
ij
(x)
x
i
x
j
,= 0 (3.3)
on o. This is a condition for the given equation and for the given surface o.
Denition. The dierential equation
n
i,j=1
a
ij
(x)
x
i
x
j
= 0
is called characteristic dierential equation associated to the given dieren-
tial equation (3.1).
If , ,= 0, is a solution of the characteristic dierential equation, then
the surface dened by = 0 is called characteristic surface.
Remark. The condition (3.3) is satised for each with ,= 0 if the
quadratic matrix (a
ij
(x)) is positive or negative denite for each x ,
which is equivalent to the property that all eigenvalues are dierent from
zero and have the same sign. This follows since there is a (x) > 0 such
that, in the case that the matrix (a
ij
) is poitive denite,
n
i,j=1
a
ij
(x)
i
j
(x)[[
2
for all R
n
. Here and in the following we assume that the matrix (a
ij
) is
real and symmetric.
The characterization of dierential equation (3.1) follows from the signs of
the eigenvalues of (a
ij
(x)).
Denition. Dierential equation (3.1) is said to be of type (, , ) at
x if eigenvalues of (a
ij
)(x) are positive, eigenvalues are negative
and eigenvalues are zero ( + + = n).
In particular, equation is called
elliptic if it is of type (n, 0, 0) or of type (0, n, 0), i. e., all eigenvalues are
dierent from zero and have the same sign,
parabolic if it is of type (n1, 0, 1) or of type (0, n1, 1), i. e., one eigenvalue
is zero and all the others are dierent from zero and have the same sign,
hyperbolic if it is of type (n 1, 1, 0) or of type (1, n 1, 0), i. e., all
eigenvalues are dierent from zero and one eigenvalue has another sign than
all the others.
Remarks:
1. According to this denition there are other types aside from elliptic,
parabolic or hyperbolic equations.
2. The classication depends in general on x . An example is the
Tricomi equation, which appears in the theory of transsonic ows,
yu
xx
+u
yy
= 0.
This equation is elliptic if y > 0, parabolic if y = 0 and hyperbolic for y < 0.
Examples:
1. The Laplace equation in R
3
is u = 0, where
u := u
xx
+u
yy
+u
zz
.
This equation is elliptic. Thus for each manifold o given by (x, y, z) :
(x, y, z) = 0, where is an arbitrary suciently regular function such
that ,= 0, all derivatives of u are known on o, provided u and u are
known on o.
2. The wave equation u
tt
= u
xx
+ u
yy
+ u
zz
, where u = u(x, y, z, t), is
hyperbolic. Such a type describes oscillations of mechanical structures, for
example.
3. The heat equation u
t
= u
xx
+u
yy
+u
zz
, where u = u(x, y, z, t), is parabolic.
It describes, for example, the propagation of heat in a domain.
4. Consider the case that the (real) coecients a
ij
in equation (3.1) are
constant. We recall that the matrix A = (a
ij
) is symmetric, i. e., A
T
= A.
In this case, the transform to principle axis leads to a normal form from
which the classication of the equation is obviously. Let U be the associated
orthogonal matrix, then
U
T
AU =
_
_
_
_
1
0 0
0
2
0
. . . . . . . . . . . . . . . .
0 0
n
_
_
_
_
.
Here is U = (z
1
, . . . , z
n
), where z
l
, l = 1, . . . , n, is an orthonormal system of
eigenvectors to the eigenvalues
l
.
Set y = U
T
x and v(y) = u(Uy), then
n
i,j=1
a
ij
u
x
i
x
j
=
n
i=1
i
v
y
i
y
j
. (3.4)
3.1.1 Normal form in two variables
Consider the dierential equation
a(x, y)u
xx
+ 2b(x, y)u
xy
+c(x, y)u
yy
+ terms of lower order = 0 (3.5)
in R
2
. The associated characteristic dierential equation is
a
2
x
+ 2b
x
y
+c
2
y
= 0. (3.6)
We show that an appropriate coordinate transform will simplify equation (3.5)
sometimes in such a way that we can solve the transformed equation explic-
itly.
Let z = (x, y) be a solution of (3.6). Consider the level sets (x, y) :
(x, y) = const. and assume
y
,= 0 at a point (x
0
, y
0
) of the level set.
Then there is a function y(x) dened in a neighbourhood of x
0
such that
(x, y(x)) = const. It follows
y
(x) =
y
,
which implies, see the characteristic equation (3.6),
ay
2
2by
+c = 0. (3.7)
Then, provided a ,= 0, we can calculate := y
from the (known) coecients

a, b and c:
1,2
=
1
a
_
b
_
b
2
ac
_
. (3.8)
These solutions are real if and only of ac b
2
0.
Equation (3.5) is hyperbolic if ac b
2
< 0, parabolic if ac b
2
= 0 and
elliptic if acb
2
> 0. This follows from an easy discussion of the eigenvalues
of the matrix
_
a b
b c
_
,
see an exercise.
Normal form of a hyperbolic equation
Let and are solutions of the characteristic equation (3.6) such that
y
1

1
=
y
y
2

2
=
y
,
where
1
and
2
are given by (3.8). Thus and are solutions of the linear
homogeneous equations of rst order
x
+
1
(x, y)
y
= 0 (3.9)
x
+
2
(x, y)
y
= 0. (3.10)
Assume (x, y), (x, y) are solutions such that ,= 0 and ,= 0, see an
exercise for the existence of such solutions.
Consider two families of level sets dened by (x, y) = and (x, y) = ,
see Figure 3.3.
y
x
(x,y)=
(x,y)=
(x,y)=
(x,y)=

1
2
1
2
Figure 3.3: Level sets
These level sets are characteristic curves of the partial dierential equa-
tions (3.9) and (3.10), respectively, see an exercise of the previous chapter.
Lemma. (i) Curves from dierent families can not touch each other.
(ii)
x
y

y
x
,= 0.
Proof. (i):
y
2
y
1

2

1
=
2
a
_
b
2
ac ,= 0.
(ii):
2

1
=

x
y
.
2
Proposition 3.1. The mapping = (x, y), = (x, y) transforms equa-
tion (3.5) into
v
= lower order terms, (3.11)

where v(, ) = u(x(, ), y(, )).
Proof. The proof follows from a straightforward calculation.
u
x
= v
x
+v
x
u
y
= v
y
+v
y
u
xx
= v
2
x
+ 2v
x
+v
2
x
+ lower order terms
u
xy
= v
y
+v
(
x
y
+
y
x
) +v
y
+ lower order terms
u
yy
= v
2
y
+ 2v
y
+v
2
y
+ lower order terms.
Thus
au
xx
+ 2bu
xy
+cu
yy
= v
+ 2v
+v
+l.o.t.,
where
: = a
2
x
+ 2b
x
y
+c
2
y
: = a
x
x
+b(
x
y
+
y
x
) +c
y
y
: = a
2
x
+ 2b
x
y
+c
2
y
.
The coecients and are zero since and are solutions of the charac-
teristic equation. Since

2
= (ac b
2
)(
x
y

y
x
)
2
,
it follows from the above lemma that the coecient is dierent from zero.
2
Example: Consider the dierential equation
u
xx
u
yy
= 0.
The associated characteristic dierential equation is
2
x

2
y
= 0.
Since
1
= 1 and
2
= 1, the functions and satisfy dierential equa-
tions
x
+
y
= 0
x

y
= 0.
Solutions with ,= 0 and ,= 0 are
= x y, = x +y.
Thus the mapping
= x y, = x +y
leads to the simple equation
v
(, ) = 0.
Assume v C
2
is a solution, then v
= f
1
() for an arbitrary C
1
function
f
1
(). It follows
v(, ) =
_

0
f
1
() d +g(),
where g is an arbitrary C
2
function. Thus each C
2
-solution of the dieren-
tial equation can be written as
() v(, ) = f() +g(),
where f, g C
2
. On the other hand, for arbitrary C
2
-functions f, g the
function () is a solution of the dierential equation v
= 0. Consequently
each C
2
-solution of the original equation u
xx
u
yy
= 0 is given by
u(x, y) = f(x y) +g(x +y),
where f, g C
2
.
3.2. QUASILINEAR EQUATIONS OF SECOND ORDER 73
3.2 Quasilinear equations of second order
Here we consider the equation
n
i,j=1
a
ij
(x, u, u)u
x
i
x
j
+b(x, u, u) = 0 (3.12)
in a domain R
n
, where u : R. We assume that a
ij
= a
ji
.
As in the previous section we can derive the characteristic equation
n
i,j=1
a
ij
(x, u, u)
x
i
x
j
= 0.
In contrast to linear equations, solutions of the characteristic equation de-
pend on the solution considered.
3.2.1 Quasilinear elliptic equations
There is a large class of quasilinear equations such that the associated char-
acteristic equation has no solution , ,= 0.
Set
U = (x, z, p) : x , z R, p R
n
.
Denition. The quasilinear equation (3.12) is called elliptic if the matrix
(a
ij
(x, z, p)) is positive denite for each (x, z, p) U.
Assume equation (3.12) is elliptic and let (x, z, p) be the minimum and
(x, z, p) the maximum of the eigenvalues of (a
ij
), then
0 < (x, z, p)[[
2
i,j=1
a
ij
(x, z, p)
i
j
(x, z, p)[[
2
for all R
n
.
Denition. Equation (3.12) is called uniformly elliptic if / is uniformly
bounded in U.
An important class of elliptic equations which are not uniformly elliptic
(nonuniformly elliptic) is
n
i=1
x
i
_
u
x
i
_
1 +[u[
2
_
+ lower order terms = 0. (3.13)
The main part is the minimal surface operator (left hand side of the minimal
surface equation). The coecients a
ij
are
a
ij
(x, z, p) =
_
1 +[p[
2
_
1/2
_
ij

p
i
p
j
1 +[p[
2
_
,
ij
denotes the Kronecker delta symbol. It follows that
=
1
(1 +[p[
2
)
3/2
, =
1
(1 +[p[
2
)
1/2
.
Thus equation (3.13) is not uniformly elliptic.
The behaviour of solutions of uniformly elliptic equations is similar to
linear elliptic equations in contrast to the behaviour of solutions of nonuni-
formly elliptic equations. Typical examples for nonuniformly elliptic equa-
tions are the minimal surface equation and the capillary equation.
3.3 Systems of rst order
Consider the quasilinear system
n
k=1
A
k
(x, u)u
u
k
+b(x, u) = 0, (3.14)
where A
k
are m m-matrices, suciently regular with respect to their ar-
guments, and
u =
_
_
_
u
1
.
.
.
u
m
_
_
_
, u
x
k
=
_
_
_
u
1,x
k
.
.
.
u
m,x
k
_
_
_
, b =
_
_
_
b
1
.
.
.
b
m
_
_
_
.
We ask the same question as above: can we calculate all derivatives of u
in a neighbourhood of a given hypersurface o in R
n
dened by (x) = 0,
,= 0, provided u(x) is given on o?
For an answer we map o onto a at surface o
0
by using the mapping
= (x) of Section 3.1 and write equation (3.14) in new coordinates. Set
v() = u(x()), then
n
k=1
A
k
(x, u)
x
k
v
n
= terms known on o
0
.
3.3. SYSTEMS OF FIRST ORDER 75
We can solve this system with respect to v
n
, provided that
det
_
n
k=1
A
k
(x, u)
x
k
_
,= 0
on o.
Denition. Equation
det
_
n
k=1
A
k
(x, u)
x
k
_
= 0
is called characteristic equation associated to equation (3.14) and a surface
o: (x) = 0, dened by a solution , ,= 0, of this characteristic equation
is said to be characteristic surface.
Set
C(x, u, ) = det
_
n
k=1
A
k
(x, u)
k
_
for R
n
.
Denition. (i) The system (3.14) is hyperbolic at (x, u(x)) if there is a
regular linear mapping = Q, where = (
1
, . . . ,
n1
, ), such that there
exists m real roots
k
=
k
(x, u(x),
1
, . . . ,
n1
), k = 1, . . . , m, of
D(x, u(x),
1
, . . . ,
n1
, ) = 0
for all (
1
, . . . ,
n1
), where
D(x, u(x),
1
, . . . ,
n1
, ) = C(x, u(x), x, Q).
(ii) System (3.14) is parabolic if there exists a regular linear mapping = Q
such that D is independent of , i. e., D depends on less than n parameters.
(iii) System (3.14) is elliptic if C(x, u, ) = 0 only if = 0.
Remark. In the elliptic case all derivatives of the solution can be calculated
from the given data and the given equation.
3.3.1 Examples
1. Beltrami equations
Wu
x
bv
x
cv
y
= 0 (3.15)
Wu
y
+av
x
+bv
y
= 0, (3.16)
where W, a, b, c are given functions depending of (x, y), W ,= 0 and the
matrix
_
a b
b c
_
is positive denite.
The Beltrami system is a generalization of Cauchy-Riemann equations.
The function f(z) = u(x, y) + iv(x, y), where z = x + iy, is called a qua-
siconform mapping, see for example [9], Chapter 12, for an application to
partial dierential equations.
Set
A
1
=
_
W b
0 a
_
, A
2
=
_
0 c
W b
_
.
Then the system (3.15), (3.16) can be written as
A
1
_
u
x
v
x
_
+A
2
_
u
y
v
y
_
=
_
0
0
_
.
Thus,
C(x, y, ) =
W
1
b
1
c
2
W
2
a
1
+b
2
= W(a
2
1
+ 2b
1
2
+c
2
2
),
which is dierent from zero if ,= 0 according to the above assumptions.
Thus the Beltrami system is elliptic.
2. Maxwell equations
The Maxwell equations in the isotropic case are
c rot
x
H = E +E
t
(3.17)
c rot
x
E = H
t
, (3.18)
where
E = (e
1
, e
2
, e
3
)
T
electric eld strength, e
i
= e
i
(x, t), x = (x
1
, x
2
, x
3
),
H = (h
1
, h
2
, h
3
)
T
magnetic eld strength, h
i
= h
i
(x, t),
c speed of light,
specic conductivity,
dielectricity constant,
magnetic permeability.
Here c, , and are positive constants.
Set p
0
=
t
, p
i
=
x
i
, i = 1, . . . 3, then the characteristic dierential equa-
tion is
p
0
/c 0 0 0 p
3
p
2
0 p
0
/c 0 p
3
0 p
1
0 0 p
0
/c p
2
p
1
0
0 p
3
p
2
p
0
/c 0 0
p
3
0 p
1
0 p
0
/c 0
p
2
p
1
0 0 0 p
0
/c
= 0.
The following manipulations simplies this equation:
(i) multiply the rst three columns with p
0
/c,
(ii) multiply the 5th column with p
3
and the the 6th column with p
2
and
add the sum to the 1st column,
(iii) multiply the 4th column with p
3
and the 6th column with p
1
and add
the sum to the 2th column,
(iv) multiply the 4th column with p
2
and the 5th column with p
1
and add
the sum to the 3th column,
(v) expand the resulting determinant with respect to the elements of the
6th, 5th and 4th row.
We obtain
q +p
2
1
p
1
p
2
p
1
p
3
p
1
p
2
q +p
2
2
p
2
p
3
p
1
p
3
p
2
p
3
q +p
2
3
= 0,
where
q :=

c
2
p
2
0
g
2
with g
2
:= p
2
1
+ p
2
2
+ p
2
3
. The evaluation of the above equation leads to
q
2
(q +g
2
) = 0, i. e.,
2
t
_
c
2
2
t
[
x
[
2
_
= 0.
It follows immediately that Maxwell equations are a hyperbolic system, see an
exercise. There are two solutions of this characteristic equation. The rst
one are characteristic surfaces o(t), dened by (x, t) = 0, which satisfy
t
= 0. These surfaces are called stationary waves. The second type of
characteristic surfaces are dened by solutions of
c
2
2
t
= [
x
[
2
.
Functions dened by = f(n xV t) are solutions of this equation. Here is
f(s) an arbitrary function with f
(s) ,= 0, n is a unit vector and V = c/
.
The associated characteristic surfaces o(t) are dened by
(x, t) f(n x V t) = 0,
here we assume that 0 is in he range of f : R R. Thus, o(t) is dened
by n xV t = c, where c is a xed constant. It follows that the planes o(t)
with normal n move with speed V in direction of n, see Figure 3.4
x
x
2
1
n
S(t)
S(0)
d(t)
Figure 3.4: d
(t) is the speed of plane waves

V is called speed of the plane wave o(t).
Remark. According to the previous discussions, singularities of a solution
of Maxwell equations are located at most on characteristic surfaces.
A special case of Maxwell equations are the telegraph equations, which
follow from Maxwell equations if div E = 0 and div H = 0, i. e., E and
H are elds free of sources. In fact, it is sucient to assume that this
assumption is satised at a xed time t
0
only, see an exercise.
Since
rot
x
rot
x
A = grad
x
div
x
A
x
A
for each C
2
-vector eld A, it follows from Maxwell equations the uncoupled
system
x
E =

c
2
E
tt
+

c
2
E
t
x
H =

c
2
H
tt
+

c
2
H
t
.
3. Equations of gas dynamics
Consider the following quasilinear equations of rst order.
v
t
+ (v
x
) v +
1
x
p = f (Euler equations).
Here is
v = (v
1
, v
2
, v
3
) the vector of speed, v
i
= v
i
(x, t), x = (x
1
, x
2
, x
3
),
p pressure, p = (x, t),
density, = (x, t),
f = (f
1
, f
2
, f
3
) density of the external force, f
i
= f
i
(x, t),
(v
x
)v (v
x
v
1
, v
x
v
2
, v
x
v
3
))
T
.
The second equation is
t
+v
x
+ div
x
v = 0 (conservation of mass).
Assume the gas is compressible and that there is a function (state equation)
p = p(),
where p
() > 0 if > 0. Then the above system of four equations is

v
t
+ (v )v +
1
() = f (3.19)
t
+ div v +v = 0, (3.20)
where
x
and div div
x
, i. e., these operators apply on the spatial
variables only.
The characteristic dierential equation is here
d
dt
0 0
1
x
1
0
d
dt
0
1
x
2
0 0
d
dt
1
x
3
x
1

x
2

x
3
d
dt
= 0,
where
d
dt
:=
t
+ (
x
) v.
Evaluating the determinant, we get the characteristic dierential equation
_
d
dt
_
2
_
_
d
dt
_
2
p
()[
x
[
2
_
= 0. (3.21)
This equation implies consequences for the speed of the characteristic sur-
faces as the following consideration shows.
Consider a family o(t) of surfaces in R
3
dened by (x, t) = c, where
x R
3
and c is a xed constant. As usually, we assume that
x
,= 0. One
of the two normals on o(t) at a point of the surface o(t) is given by, see an
exercise,
n =

x
[
x
[
. (3.22)
Let Q
0
o(t
0
) and let Q
1
o(t
1
) be a point on the line dened by Q
0
+sn,
where n is the normal (3.22 on o(t
0
) at Q
0
and t
0
< t
1
, t
1
t
0
small, see
Figure 3.5.
)
0
S(t
S(t )
1
n
Q
Q
1
0
Figure 3.5: Denition of the speed of a surface
Denition. The limit
P = lim
t
1
t
0
[Q
1
Q
0
[
t
1
t
0
is called speed of the surface o(t).
Proposition 3.2. The speed of the surface o(t) is
P =

t
[
x
[
. (3.23)
Proof. The proof follows from (Q
0
, t
0
) = 0 and (Q
0
+ dn, t
0
+ t) = 0,
where d = [Q
1
Q
0
[ and t = t
1
t
0
.
2
Set v
n
:= v n which is the component of the velocity vector in direction
n. From (3.22) we get
v
n
=
1
[
x
[
v
x
.
Denition. V := P v
n
, the dierence of the speed of the surface and the
speed of liquid particles, is called relative speed.
n
v
S
Figure 3.6: Denition of relative speed
Using the above formulas for P and v
n
it follows
V = P v
n
=

t
[
x
[

v
x
[
x
[
=
1
[
x
[
d
dt
.
Then, we obtain from the characteristic equation (3.21) that
V
2
[
x
[
2
_
V
2
[
x
[
2
p
()[
x
[
2
_
= 0.
An interesting conclusion is that there are two relative speeds: V = 0 or
V
2
= p
().
Denition.
_
p
() is called speed of sound .

3.4 Systems of second order
Here we consider the system
n
k,l=1
A
kl
(x, u, u)u
x
k
x
l
+ lower order terms = 0, (3.24)
where A
kl
are (mm) matrices and u = (u
1
, . . . , u
m
)
T
. We assume A
kl
=
A
lk
, which is no restriction of generality provided u C
2
is satised. As in
the previous sections, the classication follows from the question whether or
not we can calculate formally the solution from the dierential equations,
if suciently many data are given on an initial manifold. Let the initial
manifold o be given by (x) = 0 and assume that ,= 0. The mapping
x = x(), see previous sections, leads to
n
k,l=1
A
kl
x
k
x
l
v
n
= terms known on o,
where v() = u(x()).
The characteristic equation is here
det
_
_
n
k,l=1
A
kl
x
k
x
l
_
_
= 0.
If there is a solution with ,= 0, then it is possible that second derivatives
are not continuous in a neighbourhood of o.
Denition. The system is called elliptic if
det
_
_
n
k,l=1
A
kl
l
_
_
,= 0
for all R
n
, ,= 0.
3.4. SYSTEMS OF SECOND ORDER 83
3.4.1 Examples
1. Navier-Stokes equations
The Navier-Stokes system for a viscous incompressible liquid is
v
t
+ (v
x
)v =
1
x
p +
x
v
div
x
v = 0,
where is the (constant and positive) density of liquid,
is the (constant and positive) viscosity of liquid,
v = v(x, t) velocity vector of liquid particles, x R
3
or in R
2
,
p = p(x, t) pressure.
The problem is to nd solutions v, p of the above system.
2. Linear elasticity
Consider the system
2
u
t
2
=
x
u + ( +)
x
(div
x
u) +f. (3.25)
Here is, in the case of an elastic body in R
3
,
u(x, t) = (u
1
(x, t), u
2
(x, t), u
3
(x, t)) displacement vector,
f(x, t) density of external force,
(constant) density,
, (positive) Lame constants.
The characteristic equation is det C = 0, where the entries of the matrix
C are given by
c
ij
= ( +)
x
i
x
j
+
ij
_
[
x
[
2
2
t
_
.
The characteristic equation is
_
( + 2)[
x
[
2
2
t
_ _
[
x
[
2
2
t
_
2
= 0.
It follows that two dierent speeds P of characteristic surfaces o(t), dened
by (x, t) = const., are possible, namely
P
1
=
+ 2
, and P
2
=
_
.
We recall that P =
t
/[
x
[.
3.5 Theorem of Cauchy-Kovalevskaya
Consider the quasilinear system of rst order (3.14) of Section 3.3. Assume
an initial manifolds o is given by (x) = 0, ,= 0, and suppose that is
not characteristic. Then, see Section 3.3, the system (3.14) can be written
as
u
x
n
=
n1
i=1
a
i
(x, u)u
x
i
+b(x, u) (3.26)
u(x
1
, . . . , x
n1
, 0) = f(x
1
, . . . , x
n1
) (3.27)
Here is u = (u
1
, . . . , u
m
)
T
, b = (b
1
, . . . , b
n
)
T
and a
i
are (mm)-matrices. We
assume a
i
, b and f are in C
with respect to their arguments. From (3.26)

and (3.27) it follows that we can calculate formally all derivatives D
u in a
neigbourhood of the plane x : x
n
= 0, in particular in a neighbourhood
of 0 R
n
. Thus we have a formal power series of u(x) at x = 0:
u(x)
1
!
D
u(0)x
.
For notations and denitions used here and in the following see the appendix
to this section.
Then, as usually, two questions arise:
(i) Does the power series converge in a neighbourhood of 0 R
n
?
(ii) Is a convergent power series a solution of the initial value problem (3.26),
(3.27)?
Remark. Quite dierent to this power series method is the method of
asymptotic expansions. Here one is interested in a good approximation of
an unknown solution of an equation by a nite sum
N
i=0
i
(x) of functions
i
. In general, the innite sum
i=0
i
(x) does not converge, in contrast
to the power series method of this section. See [15] for some asymptotic
formulas in capillarity.
Theorem 3.1 (Cauchy-Kovalevskaya). There is a neighbourhood of 0 R
n
such there is a real analytic solution of the initial value problem (3.26),
(3.27). This solution is unique in the class of real analytic functions.
3.5. THEOREM OF CAUCHY-KOVALEVSKAYA 85
Proof. The proof is taken from F. John [10]. We introduce uf as the new
solution for which we are looking at and we add a new coordinate u
to the
solution vector by setting u
(x) = x
n
. Then
u
x
n
= 1, u
x
k
= 0, k = 1, . . . , n 1, u
(x
1
, . . . , x
n1
, 0) = 0
and the extended system (3.26), (3.27) is
_
_
_
_
_
u
1,x
n
.
.
.
u
m,x
n
u
x
n
_
_
_
_
_
=
n1
i=1
_
a
i
0
0 0
_
_
_
_
_
_
u
1,x
i
.
.
.
u
m,x
i
u
x
i
_
_
_
_
_
+
_
_
_
_
_
b
1
.
.
.
b
m
1
_
_
_
_
_
,
where the associated initial condition is u(x
1
, . . . , x
n1
, 0) = 0. The new u
is u = (u
1
, . . . , u
m
)
T
, the new a
i
are a
i
(x
1
, . . . , x
n1
, u
1
, . . . , u
m
, u
) and the
new b is b = (x
1
, . . . , x
n1
, u
1
, . . . , u
m
, u
)
T
.
Thus we are led to an initial value problem of the type
u
j,x
n
=
n1
i=1
N
k=1
a
i
jk
(z)u
k,x
i
+b
j
(z), j = 1, . . . , N (3.28)
u
j
(x) = 0 if x
n
= 0, (3.29)
where j = 1, . . . , N and z = (x
1
, . . . , x
n1
, u
1
, . . . , u
N
).
The point here is that a
i
jk
and b
j
are independent of x
n
. This fact
simplies the proof of the theorem.
From (3.28) and (3.29) we can calculate formally all D
u
j
. Then we
have formal power series for u
j
:
u
j
(x)
c
(j)
,
where
c
(j)
=
1
!
D
u
j
(0).
We will show that these power series are (absolutely) convergent in a neigh-
bourhood of 0 R
n
, i. e., they are real analytic functions, see the appendix
for the denition of real analytic functions. Inserting these functions into
the left and into the right hand side of (3.28) we obtain on the right and on
the left hand side real analytic functions. This follows since compositions
of real analytic functions are real analytic again, see Proposition A7 of the
appendix to this section. The resulting power series on the left and on the
right have the same coecients caused by the calculation of the derivatives
D
u
j
(0) from (3.28). It follows that u
j
(x), j = 1, . . . , n, dened by its
formal power series are solutions of the initial value problem (3.28), (3.29).
Set
d =
_

z
1
, . . . ,

z
N+n1
_
Lemma A. Assume u C
in a neighbourhood of 0 R
n
. Then
D
u
j
(0) = P
_
d
a
i
jk
(0), d
b
j
(0)
_
,
where [[, [[ [[ and P
are polynomials in the indicated arguments with

nonnegative integers as coecients which are independent of a
i
and of
b.
Proof. It follows from equation (3.28) that
D
n
D
u
j
(0) = P
(d
a
i
jk
(0), d
b
j
(0), D
u
k
(0)). (3.30)
Here is D
n
= /x
n
and , , , satisfy the inequalities
[[, [[ [[, [[ [[ + 1,
and, which is essential in the proof, the last coordinates in the multi-indices
= (
1
, . . . ,
n
), = (
1
, . . . ,
n
) satisfy
n

n
since the right hand side
of (3.28) is independent of x
n
. Moreover, it follows from (3.28) that the
polynomials P
have integers as coecients. The initial condition (3.29)

implies
D
u
j
(0) = 0, (3.31)
where = (
1
, . . . ,
n1
, 0), that is,
n
= 0. Then, the proof is by induction
with respect to
n
. The induction starts with
n
= 0, then we replace
D
u
k
(0) in the right hand side of (3.30) by (3.31), that is by zero. Then it
follows from (3.30) that
D
u
j
(0) = P
(d
a
i
jk
(0), d
b
j
(0), D
u
k
(0)),
where = (
1
, . . . ,
n1
, 1). 2
Denition. Let f = (f
1
, . . . , f
m
), F = (F
1
, . . . , F
m
), f
i
= f
i
(x), F
i
= F
i
(x),
and f, F C
. We say f is majorized by F if
[D
f
k
(0)[ D
F
k
(0), k = 1, . . . , m
for all . We write f << F, if f is majorized by F.
Denition. The initial value problem
U
j,x
n
=
n1
i=1
N
k=1
A
i
jk
(z)U
k,x
i
+B
j
(z) (3.32)
U
j
(x) = 0 if x
n
= 0, (3.33)
j = 1, . . . , N, A
i
jk
, B
j
real analytic, ist called majorizing problem to (3.28),
(3.29) if
a
i
jk
<< A
i
jk
and b
j
<< B
j
.
Lemma B. The formal power series
1
!
D
u
j
(0)x
,
where D
u
j
(0) are dened in Lemma A, is convergent in a neighbourhood of
0 R
n
if there exists a majorizing problem which has a real analytic solution
U in x = 0, and
[D
u
j
(0)[ D
U
j
(0).
Proof. It follows from Lemma A and from the assumption of Lemma B that
[D
u
j
(0)[ P
_
[d
a
i
jk
(0)[, [d
b
j
(0)[
_
P
_
[d
A
i
jk
(0)[, [d
B
j
(0)[
_
D
U
j
(0).
The formal power series
1
!
D
u
j
(0)x
,
is convergent since
1
!
[D
u
j
(0)x
1
!
D
U
j
(0)[x
[.
The right hand side is convergent in a neighbourhood of x R
n
by assump-
tion. 2
Lemma C. There is a majorising problem which has a real analytic solution.
Proof. Since a
i
ij
(z), b
j
(z) are real analytic in a neighbourhood of z = 0 it
follows from Proposition A5 of the appendix to this section that there are
positive constants M and r such that all these functions are majorized by
Mr
r z
1
. . . z
N+n1
.
Thus a majorizing problem is
U
j,x
n
=
Mr
r x
1
. . . x
n1
U
1
. . . U
N
_
1 +
n1
i=1
N
k=1
U
k,x
i
_
U
j
(x) = 0 if x
n
= 0,
j = 1, . . . , N.
The solution of this problem is
U
j
(x
1
, . . . , x
n1
, x
n
) = V (x
1
+. . . +x
n1
, x
n
), j = 1, . . . , N,
where V (s, t), s = x
1
+ . . . + x
n1
, t = x
n
, is the solution of the Cauchy
initial value problem
V
t
=
Mr
r s NV
(1 +N(n 1)V
s
) ,
V (s, 0) = 0.
which has the solution, see an exercise,
V (s, t) =
1
Nn
_
r s
_
(r s)
2
2nMNrt
_
.
This function is real analytic in (s, t) at (0, 0). It follows that U
j
(x) are also
real analytic functions. Thus the Cauchy-Kovalevskaya theorem is shown.
2
Examples:
1. Ordinary dierential equations
Consider the initial value problem
y
(x) = f(x, y(x))

y(x
0
) = y
0
,
where x
0
R and y
0
R
n
are given. Assume f(x, y) is real analytic in a
neighbourhood of (x
0
, y
0
) RR
n
. Then it follows from the above theorem
that there exists an analytic solution y(x) of the initial value problem in a
neighbourhood of x
0
. This solution is unique in the class of analytic func-
tions according to the theorem of Cauchy-Kovalevskaya. From the Picard-
Lindelof theorem it follows that this analytic solution is unique even in the
class of C
1
-functions.
2. Partial dierential equations of second order
Consider the boundary value problem for two variables
u
yy
= f(x, y, u, u
x
, u
y
, u
xx
, u
xy
)
u(x, 0) = (x)
u
y
(x, 0) = (x).
We assume that , are analytic in a neighbourhood of x = 0 and that f
is real analytic in a neighbourhood of
(0, 0, (0),
(0), (0),
(0)).
There exists a real analytic solution in a neigbourhood of 0 R
2
of the above
initial value problem.
In particular, there is a real analytic solution in a neigbourhood of 0 R
2
of the initial value problem
u = 1
u(x, 0) = 0
u
y
(x, 0) = 0.
The proof follows by writing the above problem as a system. Set p = u
x
,
q = u
y
, r = u
xx
, s = u
xy
, t = u
yy
, then
t = f(x, y, u, p, q, r, s).
Set U = (u, p, q, r, s, t)
T
, b = (q, 0, t, 0, 0, f
y
+f
u
q +f
q
t)
T
and
A =
_
_
_
_
_
_
_
_
0 0 0 0 0 0
0 0 1 0 0 0
0 0 0 0 0 0
0 0 0 0 1 0
0 0 0 0 0 1
0 0 f
p
0 f
r
f
s
_
_
_
_
_
_
_
_
.
Then the rewritten dierential equation is the system U
y
= AU
x
+ b with
the initial condition
U(x, 0) =
_
(x),
(x), (x),
(x),
(x), f
0
(x)
_
,
where f
0
(x) = f(x, 0, (x),
(x), (x),
(x),
(x)).
3.5.1 Appendix: Real analytic functions
Multi-index notation
The following multi-index notation simplies many presentations of formu-
las. Let x = (x
1
, . . . , x
n
) and
u : R
n
R (or R
m
for systems).
The n-tuple of nonnegative integers (including zero)
= (
1
, . . . ,
n
)
is called multi-index. Set
[[ =
1
+. . . +
n
! =
1
!
2
! . . .
n
!
x
= x
1
1
x
2
2
. . . x
n
n
(for a monom)
D
k
=

x
k
D = (D
1
, . . . , D
n
)
Du = (D
1
u, . . . , D
n
u) u grad u
D
= D
1
1
D
2
2
. . . D
n
n

||
x
1
1
x
2
2
. . . x
n
n
.
Dene a partial order by
if and only if
i

i
for all i.
Sometimes we use the notations
0 = (0, 0 . . . , 0), 1 = (1, 1 . . . , 1),
where 0, 1 R
n
.
Using this multi-index notion, we have
1.
(x +y)
,
+ =
!
!!
x
,
where x, y R
n
and , , are multi-indices.
2. Taylor expansion for a polynomial f(x) of degree m:
f(x) =
||m
1
!
(D
f(0)) x
,
here is D
f(0) := (D
f(x)) [
x=0
.
3. Let x = (x
1
, . . . , x
n
) and m 0 an integer, then
(x
1
+. . . +x
n
)
m
=
||=m
m!
!
x
.
4.
! [[! n
||
!.
5. Leibnizs rule:
D
(fg) =
,
+ =
!
!!
(D
f)(D
g).
6.
D
=
!
( )!
x
if ,
D
= 0 otherwise.
7. Directional derivative:
d
m
dt
m
f(x +ty) =
||=m
[[!
!
(D
f(x +ty)) y
,
where x, y R
n
and t R.
8. Taylors theorem: Let u C
m+1
in a neighbourhood N(y) of y, then, if
x N(y),
u(x) =
||m
1
!
(D
u(y)) (x y)
+R
m
,
where
R
m
=
||=m+1
1
!
(D
u(y +(x y))) x
, 0 < < 1,
= (u, m, x, y), or
R
m
=
1
m!
_
1
0
(1 t)
m
(m+1)
(t) dt,
where (t) = u(y +t(x y)). It follows from 7. that
R
m
= (m+ 1)
||=m+1
1
!
__
1
0
(1 t)D
u(y +t(x y)) dt

_
(x y)
.
9. Using multi-index notation, the general linear partial dierential equation
of order m can be written as
||m
a
(x)D
u = f(x) in R
n
.
Power series
Here we collect some denitions and results for power series in R
n
.
Denition. Let c
R (or R
m
). The series
m=0
_
_

||=m
c
_
_
is said to be convergent if
[c
m=0
_
_

||=m
[c
[
_
_
is convergent.
Remark. According to the above denition, a convergent series is abso-
lutely convergent. It follows that we can rearrange the order of summation.
Using the above multi-index notation and keeping in mind that we can
rearrange convergent series, we have
10. Let x R
n
, then
=
n
i=1
_

i
=0
x
i
i
_
=
1
(1 x
1
)(1 x
2
) . . . (1 x
n
)
=
1
(1 x)
1
,
provided [x
i
[ < 1 is satised for each i.
11. Assume x R
n
and [x
1
[ +[x
2
[ +. . . +[x
n
[ < 1, then
[[!
!
x
j=0
||=j
[[!
!
x
j=0
(x
1
+. . . +x
n
)
j
=
1
1 (x
1
+. . . +x
n
)
.
12. Let x R
n
, [x
i
[ < 1 for all i, and is a given multi-index. Then
!
( )!
x
= D
1
(1 x)
1
=
!
(1 x)
1+
.
13. Let x R
n
and [x
1
[ +. . . +[x
n
[ < 1. Then
[[!
( )!
x
= D
1
1 x
1
. . . x
n
=
[[!
(1 x
1
. . . x
n
)
1+||
.
Consider the power series
(3.34)
and assume this series is convergent for a z R
n
. Then, by denition,
:=
[c
[[z
[ <
and the series (3.34) is uniformly convergent for all x Q(z), where
Q(z) : [x
i
[ [z
i
[ for all i.
Thus the power series (3.34) denes a continuous function dened on Q(z),
according to a theorem of Weierstrass.
The interior of Q(z) is not empty if and only if z
i
,= 0 for all i, see
Figure 3.7. For given x in a xed compact subset D of Q(z) there is a
q, 0 < q < 1, such that
[x
i
[ q[z
i
[ for all i.
Set
f(x) =
.
z
Q(z)
D
Figure 3.7: Denition of D Q(z)
Proposition A1. (i) In every compact subset D of Q(z) one has f
C
(D) and the formal dierentiate series, that is
, is uniformly
convergent on the closure of D and is equal to D
f.
(ii)
[D
f(x)[ M[[!r
||
in D,
where
M =

(1 q)
n
, r = (1 q) min
i
[z
i
[.
Proof. See F. John [10], p. 64. Or an exercise. Hint: Use formula 12. where
x is replaced by (q, . . . , q).
Remark. From the proposition above it follows
c
=
1
!
D
f(0).
Denition. Assume f is dened on a domain R
n
, then f is said to be
real analytic in y if there are c
R and if there is a neighbourhood

N(y) of y such that
f(x) =
(x y)
for all x N(y), and the series converges (absolutely) for each x N(y).
A function f is called real analytic in if it is real analytic for each y .
We will write f C
() in the case that f is real analytic in the domain .

A vector valued function f(x) = (f
1
(x), . . . , f
m
) is called real analytic if
each coordinate is real analytic.
Proposition A2. (i) Let f C
(). Then f C
().
(ii) Assume f C
(). Then for each y there exists a neighbourhood

N(y) and positive constants M, r such that
f(x) =
1
!
(D
f(y))(x y)
for all x N(y), and the series converges (absolutely) for each x N(y),
and
[D
f(x)[ M[[!r
||
.
The proof follows from Proposition A1.
An open set R
n
is called connected if is not a union of two nonempty
open sets with empty intersection. An open set R
n
is connected if and
only if its path connected, see [11], pp. 38, for example. We say that
is path connected if for any x, y there is a continuous curve (t) ,
0 t 1, with (0) = x and (1) = y. From the theory of one complex
variable we know that a continuation of an analytic function is uniquely
determined. The same is true for real analytic functions.
Proposition A3. Assume f C
() and is connected. Then f is

uniquely determined if for one z all D
f(z) are known.

Proof. See F. John [10], p. 65. Suppose g, h C
() and D
g(z) = D
h(z)
for every . Set f = g h and
1
= x : D
f(x) = 0 for all ,
2
= x : D
f(x) ,= 0 for at least one .

The set
2
is open since D
f are continuous in . The set

1
is also open
since f(x) = 0 in a neighbourhood of y
1
. This follows from
f(x) =
1
!
(D
f(y))(x y)
.
Since z
1
, i. e.,
1
,= , it follows
2
= . 2
It was shown in Proposition A2 that derivatives of a real analytic function
satisfy estimates. On the other hand it follows, see the next proposition,
that a function f C
is real analytic if these estimates are satised.

Denition. Let y and M, r positive constants. Then f is said to be
in the class C
M,r
(y) if f C
in a neighbourhood of y and if
[D
f(y)[ M[[!r
||
for all .
Proposition A4. f C
() if and only if f C
() and for every

compact subset S there are positive constants M, r such that
f C
M,r
(y) for all y S.
Proof. See F. John [10], pp. 65-66. We will prove the local version of the
proposition, that is, we show it for each xed y . The general version
follows from Heine-Borel theorem. Because of Proposition A3 it remains to
show that the Taylor series
1
!
D
f(y)(x y)
converges (absolutely) in a neighbourhood of y and that this series is equal

to f(x).
Dene a neighbourhood of y by
N
d
(y) = x : [x
1
y
1
[ +. . . +[x
n
y
n
[ < d,
where d is a suciently small positive constant. Set (t) = f(y +t(x y)).
The one-dimensional Taylor theorem says
f(x) = (1) =
j1
k=0
1
k!
(k)
(0) +r
j
,
where
r
j
=
1
(j 1)!
_
1
0
(1 t)
j1
(j)
(t) dt.
From formula 7. for directional derivatives it follows for x N
d
(y) that
1
j!
d
j
dt
j
(t) =
||=j
1
!
D
f(y +t(x y))(x y)
.
From the assumption and the multinomial formula 3. we get for 0 t 1
1
j!
d
j
dt
j
(t)
||=j
[[!
!
r
||
[(x y)
[
= Mr
j
([x
1
y
1
[ +. . . +[x
n
y
n
[)
j
M
_
d
r
_
j
.
Choose d > 0 such that d < r, then the Taylor series converges (absolutely)
in N
d
(y) and it is equal to f(x) since the remainder satises, see the above
estimate,
[r
j
[ =
1
(j 1)!
_
1
0
(1 t)
j1
j
(t) dt
M
_
d
r
_
j
.
2
We recall that the notation f << F (f is majorized by F) was dened in
the previous section.
Proposition A5. (i) f = (f
1
, . . . , f
m
) C
M,r
(0) if and only if f <<
(, . . . , ), where
(x) =
Mr
r x
1
. . . x
n
.
(ii) f C
M,r
(0) and f(0) = 0 if and only if
f << ( M, . . . , M),
where
(x) =
M(x
1
+. . . +x
n
)
r x
1
. . . x
n
.
Proof.
D
(0) = M[[!r
||
.
2
Remark. The denition of f << F implies, trivially, that D
f << D
F.
The next proposition shows that compositions majorize if the involved func-
tions majorize. More precisely, we have
Proposition A6. Let f, F : R
n
R
m
and g, G maps a neighbourhood of
0 R
m
into R
p
. Assume all functions f(x), F(x), g(u), G(u) are in C
,
f(0) = F(0) = 0, f << F and g << G. Then g(f(x)) << G(F(x)).
Proof. See F. John [10], p. 68. Set
h(x) = g(f(x)), H(x) = G(F(x)).
For each coordinate h
k
of h we have, according to the chain rule,
D
h
k
(0) = P
g
l
(0), D
f
j
(0)),
where P
are polynomials with nonnegative integers as coecients, P
are
independent on g or f and := (/u
1
, . . . , /u
m
). Thus,
[D
h
k
(0)[ P
([
g
l
(0)[, [D
f
j
(0)[)
P
G
l
(0), D
F
j
(0))
= D
H
k
(0).
2
Using this result and Proposition A4, which characterizes real analytic func-
tions, it follows that compositions of real analytic functions are real analytic
functions again.
Proposition A7. Assume f(x) and g(u) are real analytic, then g(f(x)) is
real analytic if f(x) is in the domain of denition of g.
Proof. See F. John [10], p. 68. Assume that f maps a neighbourhood of
y R
n
in R
m
and g maps a neighbourhood of v = f(y) in R
m
. Then
f C
M,r
(y) and g C
,
(v) implies
h(x) := g(f(x)) C
,r/(mM+)
(y).
Once one has shown this inclusion, the proposition follows from Proposi-
tion A4. To show the inclusion, we set
h(y +x) := g(f(y +x)) g(v +f(y +x) f(x)) =: g
(f
(x)),
where v = f(y) and
g
(u) : = g(v +u) C

,
(0)
f
(x) : = f(y +x) f(y) C

M,r
(0).
In the above formulas v, y are considered as xed parameters. From Propo-
sition A5 it follows
f
(x) << ( M, . . . , M) =: F
g
(u) << (, . . . , ) =: G,
where
(x) =
Mr
r x
1
x
2
. . . x
n
(u) =

x
1
x
2
. . . x
n
.
From Proposition A6 we get
h(y +x) << ((x), . . . , (x)) G(F),
where
(x) =

m((x) M)
=
(r x
1
. . . x
n
)
r ( +mM)(x
1
+. . . +x
n
)
<<
r
r ( +mM)(x
1
+. . . +x
n
)
=
r/( +mM)
r/( +mM) (x
1
+. . . x
n
)
.
See an exercise for the <<-inequality. 2
3.6. EXERCISES 101
3.6 Exercises
1. Let : R
n
R in C
1
, ,= 0. Show that for given x
0
R
n
there
is in a neighbourhood of x
0
a local dieomorphism = (x), :
(x
1
, . . . , x
n
) (
1
, . . . ,
n
), such that
n
= (x).
2. Show that the dierential equation
a(x, y)u
xx
+ 2b(x, y)u
xy
+c(x, y)u
yy
+ lower order terms = 0
is elliptic if ac b
2
> 0, parabolic if ac b
2
= 0 and hyperbolic if
ac b
2
< 0.
3. Show that in the hyperbolic case there exists a solution of
x
+
1
y
=
0, see equation (3.9), such that ,= 0.
Hint: Consider an appropriate Cauchy initial value problem.
4. Show equation (3.4).
5. Find the type of
Lu := 2u
xx
+ 2u
xy
+ 2u
yy
= 0
and transform this equation into an equation with vanishing mixed
derivatives by using the orthogonal mapping (transform to principal
axis) x = Uy, U orthogonal.
6. Determine the type of the following equation at (x, y) = (1, 1/2).
Lu := xu
xx
+ 2yu
xy
+ 2xyu
yy
= 0.
7. Find all C
2
-solutions of
u
xx
4u
xy
+u
yy
= 0.
Hint: Transform to principal axis and stretching of axis lead to the
wave equation.
8. Oscillations of a beam are described by
w
x

1
E
t
= 0
x
w
t
= 0,
where stresses, w deection of the beam and E, are positive con-
stants.
a) Determine the type of the system.
b) Transform the system into two uncoupled equations, that is, w,
occur only in one equation, respectively.
c) Find non-zero solutions.
9. Find nontrivial solutions ( ,= 0) of the characteristic equation to
x
2
u
xx
u
yy
= f(x, y, u, u),
where f is given.
10. Determine the type of
u
xx
xu
yx
+u
yy
+ 3u
x
= 2x,
where u = u(x, y).
11. Transform equation
u
xx
+ (1 y
2
)u
xy
= 0,
u = u(x, y), into its normal form.
12. Transform the Tricomi-equation
yu
xx
+u
yy
= 0,
u = u(x, y), where y < 0, into its normal form.
13. Transform equation
x
2
u
xx
y
2
u
yy
= 0,
u = u(x, y), into its normal form.
14. Show that
=
1
(1 +[p[
2
)
3/2
, =
1
(1 +[p[
2
)
1/2
.
are the minimum and maximum of eigenvalues of the matrix (a
ij
),
where
a
ij
=
_
1 +[p[
2
_
1/2
_
ij

p
i
p
j
1 +[p[
2
_
.
15. Show that Maxwell equations are a hyperbolic system.
3.6. EXERCISES 103
16. Consider Maxwell equations and prove that div E = 0 and div H = 0
for all t if these equations are satised for a xed time t
0
.
Hint. div rot A = 0 for each C
2
-vector eld A = (A
1
, A
2
, A
3
).
17. Assume a characteristic surface o(t) in R
3
is dened by (x, y, z, t) =
const. such that
t
= 0 and
z
,= 0. Show that o(t) has a nonparamet-
ric representation z = u(x, y, t) with u
t
= 0, that is o(t) is independent
of t.
18. Prove formula (3.22) for the normal on a surface.
19. Prove formula (3.23) for the speed of the surface o(t).
20. Write the Navier-Stokes system as a system of type (3.24).
21. Show that the following system (linear elasticity, stationary case of (3.25)
in the two dimensional case) is elliptic
u + ( +) grad(div u) +f = 0,
where u = (u
1
, u
2
). The vector f = (f
1
, f
2
) is given and , are
positive constants.
22. Discuss the type of the following system in stationary gas dynamics
(isentrop ow) in R
2
.
uu
x
+vu
y
+a
2
x
= 0
uv
x
+vv
y
+a
2
y
= 0
(u
x
+v
y
) +u
x
+v
y
= 0.
Here are (u, v) velocity vector, density and a =
_
p
() the sound
velocity.
23. Show formula 7. (directional derivative).
Hint: Induction with respect to m.
24. Let y = y(x) be the solution of:
y
(x) = f(x, y(x))

y(x
0
) = y
0
,
where f is real analytic in a neighbourhood of (x
0
, y
0
) R
2
. Find the
polynomial P of degree 2 such that
y(x) = P(x x
0
) +O([x x
0
[
3
)
as x x
0
.
25. Let u be the solution of
u = 1
u(x, 0) = u
y
(x, 0) = 0.
Find the polynomial P of degree 2 such that
u(x, y) = P(x, y) +O((x
2
+y
2
)
3/2
)
as (x, y) (0, 0).
26. Solve the Cauchy initial value problem
V
t
=
Mr
r s NV
(1 +N(n 1)V
s
)
V (s, 0) = 0.
Hint: Multiply the dierential equation with (r s NV ).
27. Write
2
u = u as a system of rst order.
Hint:
2
u (u).
28. Write the minimal surface equation
x
_
_
u
x
_
1 +u
2
x
+u
2
y
_
_
+

y
_
_
u
y
_
1 +u
2
x
+u
2
y
_
_
= 0
as a system of rst order.
Hint: v
1
:= u
x
/
_
1 +u
2
x
+u
2
y
, v
2
:= u
y
/
_
1 +u
2
x
+u
2
y
.
29. Let f : R R
m
R
m
be real analytic in (x
0
, y
0
). Show that a real
analytic solution in a neighbourhood of x
0
of the problem
y
(x) = f(x, y)
y(x
0
) = y
0
exists and is equal to the unique C
1
[x
0
, x
0
+ ]-solution from the
Picard-Lindelof theorem, > 0 suciently small.
3.6. EXERCISES 105
30. Show (see the proof of Proposition A7)
(r x
1
. . . x
n
)
r ( +mM)(x
1
+. . . +x
n
)
<<
r
r ( +mM)(x
1
+. . . +x
n
)
.
Hint: Leibnizs rule.
Chapter 4
Hyperbolic equations
Here we consider hyperbolic equations of second order, mainly wave equa-
tions.
4.1 One-dimensional wave equation
The one-dimensional wave equation is given by
1
c
2
u
tt
u
xx
= 0, (4.1)
where u = u(x, t) is a scalar function of two variables and c is a positive
constant. According to previous considerations, all C
2
-solutions of the wave
equation are
u(x, t) = f(x +ct) +g(x ct), (4.2)
with arbitrary C
2
-functions f and g
The Cauchy initial value problem for the wave equation is to nd a C
2
-
solution of
1
c
2
u
tt
u
xx
= 0
u(x, 0) = (x)
u
t
(x, 0) = (x),
where , C
2
(, ) are given.
107
108 CHAPTER 4. HYPERBOLIC EQUATIONS
Theorem 4.1. There exists a unique C
2
(R R)-solution of the Cauchy
initial value problem, and this solution is given by dAlemberts
1
formula
u(x, t) =
(x +ct) +(x ct)
2
+
1
2c
_
x+ct
xct
(s) ds. (4.3)
Proof. Assume there is a solution u(x, t) of the Cauchy initial value problem,
then it follows from (4.2) that
u(x, 0) = f(x) +g(x) = (x) (4.4)
u
t
(x, 0) = cf
(x) cg
(x) = (x). (4.5)

From (4.4) we obtain
f
(x) +g
(x) =
(x),
which implies, together with (4.5), that
f
(x) =

(x) +(x)/c
2
g
(x) =

(x) (x)/c
2
.
Then
f(x) =
(x)
2
+
1
2c
_
x
0
(s) ds +C
1
g(x) =
(x)
2

1
2c
_
x
0
(s) ds +C
2
.
The constants C
1
, C
2
satisfy
C
1
+C
2
= f(x) +g(x) (x) = 0,
see (4.4). Thus each C
2
-solution of the Cauchy initial value problem is given
by dAlemberts formula. On the other hand, the function u(x, t) dened by
the right hand side of (4.3) is a solution of the initial value problem. 2
Corollaries. 1. The solution u(x, t) of the initial value problem depends
on the values of at the endpoints of the interval [x ct, x +ct] and on the
values of on this interval only, see Figure 4.1. The interval [x ct, x +ct]
is called domain of dependence.
1
dAlembert, Jean Babtiste le Rond, 1717-1783
4.2. HIGHER DIMENSIONS 109
x+ct=const.
t
x x x +ct
x ct
(x ,t )
0 0
0
0
0 0
0
Figure 4.1: Interval of dependence
2. Let P be a point on the x-axis. Then we ask which points (x, t) need
values of or at P in order to calculate u(x, t)? From the dAlembert
formula it follows that this domain is a cone, see Figure 4.2. This set is
called domain of inuence.
t
x
P
xct=const.
Figure 4.2: Domain of inuence
4.2 Higher dimensions
Set
2u = u
tt
c
2
u,
x
=
2
/x
2
1
+. . . +
2
/x
2
n
,
and consider the initial value problem
2u = 0 in R
n
R (4.6)
u(x, 0) = f(x) (4.7)
u
t
(x, 0) = g(x), (4.8)
where f and g are given C
2
(R
2
)-functions.
By using spherical means and the above dAlembert formula we will
derive a formula for the solution of this initial value problem.
Method of spherical means
Dene the spherical mean for a C
2
-solution u(x, t) of the initial value prob-
lem by
M(r, t) =
1
n
r
n1
_
B
r
(x)
u(y, t) dS
y
, (4.9)
where
n
= (2)
n/2
/(n/2)
is the area of the n-dimensional sphere,
n
r
n1
is the area of a sphere with
radius r.
From the mean value theorem of the integral calculus we obtain the
function u(x, t) for which we are looking at by
u(x, t) = lim
r0
M(r, t). (4.10)
Using the initial data, we have
M(r, 0) =
1
n
r
n1
_
B
r
(x)
f(y) dS
y
=: F(r) (4.11)
M
t
(r, 0) =
1
n
r
n1
_
B
r
(x)
g(y) dS
y
=: G(r), (4.12)
which are the spherical means of f and g.
The next step is to derive a partial dierential equation for the spherical
mean. From denition (4.9) of the spherical mean we obtain, after the
mapping = (y x)/r, where x and r are xed,
M(r, t) =
1
n
_
B
1
(0)
u(x +r, t) dS
.
It follows
M
r
(r, t) =
1
n
_
B
1
(0)
n
i=1
u
y
i
(x +r, t)
i
dS
=
1
n
r
n1
_
B
r
(x)
n
i=1
u
y
i
(y, t)
i
dS
y
.
Integration by parts yields
1
n
r
n1
_
B
r
(x)
n
i=1
u
y
i
y
i
(y, t) dy
since (y x)/r is the exterior normal at B
r
(x). Assume u is a solution
of the wave equation, then
r
n1
M
r
=
1
c
2
n
_
B
r
(x)
u
tt
(y, t) dy
=
1
c
2
n
_
r
0
_
B
c
(x)
u
tt
(y, t) dS
y
dc.
The previous equation follows by using spherical coordinates. Consequently
(r
n1
M
r
)
r
=
1
c
2
n
_
B
r
(x)
u
tt
(y, t) dS
y
=
r
n1
c
2
2
t
2
_
1
n
r
n1
_
B
r
(x)
u(y, t) dS
y
_
=
r
n1
c
2
M
tt
.
Thus we arrive at the dierential equation
(r
n1
M
r
)
r
= c
2
r
n1
M
tt
,
which can be written as
M
rr
+
n 1
r
M
r
= c
2
M
tt
. (4.13)
This equation (4.13) is called Euler-Poisson-Darboux equation.
4.2.1 Case n=3
The Euler-Poisson-Darboux equation in this case is
(rM)
rr
= c
2
(rM)
tt
.
Thus rM is the solution of the one-dimensional wave equation with initial
data
(rM)(r, 0) = rF(r) (rM)
t
(r, 0) = rG(r). (4.14)
From the dAlembert formula we get formally
M(r, t) =
(r +ct)F(r +ct) + (r ct)F(r ct)
2r
+
1
2cr
_
r+ct
rct
G() d. (4.15)
The right hand side of the previous formula is well dened if the domain
of dependence [x ct, x + ct] is a subset of (0, ). We can extend F and
G to F
0
and G
0
which are dened on (, ) such that rF
0
and rG
0
are
C
2
(R)-functions as follows. Set
F
0
(r) =
_
_
_
F(r) : r > 0
f(x) : r = 0
F(r) : r < 0
The function G
0
(r) is given by the same denition where F and f are re-
placed by G and g, respectively.
Lemma. rF
0
(r), rG
0
(r) C
2
(R
2
).
Proof. From denition of F(r) and G(r), r > 0, it follows from the mean
value theorem
lim
r+0
F(r) = f(x), lim
r+0
G(r) = g(x).
Thus rF
0
(r) and rG
0
(r) are C(R)-functions. These functions are also in
C
1
(R). This follows since F
0
and G
0
are in C
1
(R). We have, for example,
F
(r) =
1
n
_
B
1
(0)
n
j=1
f
y
j
(x +r)
j
dS
(+0) =
1
n
_
B
1
(0)
n
j=1
f
y
j
(x)
j
dS
=
1
n
n
j=1
f
y
j
(x)
_
B
1
(0)
n
j
dS
= 0.
Then, rF
0
(r) and rG
0
(r) are in C
2
(R), provided F
and G
are bounded as
r +0. This property follows from
F
(r) =
1
n
_
B
1
(0)
n
i,j=1
f
y
i
y
j
(x +r)
i
j
dS
.
Thus
F
(+0) =
1
n
n
i,j=1
f
y
i
y
j
(x)
_
B
1
(0)
n
i
n
j
dS
.
We recall that f, g C
2
(R
2
) by assumption. 2
The solution of the above initial value problem, where F and G are replaced
by F
0
and G
0
, respectively, is
M
0
(r, t) =
(r +ct)F
0
(r +ct) + (r ct)F
0
(r ct)
2r
+
1
2cr
_
r+ct
rct
G
0
() d.
Since F
0
and G
0
are even functions, we have
_
ctr
rct
G
0
() d = 0.
Thus
M
0
(r, t) =
(r +ct)F
0
(r +ct) (ct r)F
0
(ct r)
2r
+
1
2cr
_
ct+r
ctr
G
0
() d, (4.16)
ct+r
rct ctr
Figure 4.3: Changed domain of integration
see Figure 4.3. For xed t > 0 and 0 < r < ct it follows that M
0
(r, t) is the
solution of the initial value problem with given initially data (4.14) since
F
0
(s) = F(s), G
0
(s) = G(s) if s > 0. Since for xed t > 0
u(x, t) = lim
r0
M
0
(r, t),
it follows from dHospitals rule that
u(x, t) = ctF
(ct) +F(ct) +tG(ct)

=
d
dt
(tF(ct)) +tG(ct).
Theorem 4.2. Assume f C
3
(R
3
) and g C
2
(R
3
) are given. Then
there exists a unique solution u C
2
(R
3
[0, )) of the initial value prob-
lem (4.6)-(4.7), where n = 3, and the solution is given by the Poissons
formula
u(x, t) =
1
4c
2
t
_
1
t
_
B
ct
(x)
f(y) dS
y
_
+
1
4c
2
t
_
B
ct
(x)
g(y) dS
y
. (4.17)
Proof. Above we have shown that a C
2
-solution is given by Poissons for-
mula. Under the additional assumption f C
3
it follows from Poissons
formula that this formula denes a solution which is in C
2
, see F. John [10],
p. 129. 2
Corollary. From Poissons formula we see that the domain of dependence
for u(x, t
0
) is the intersection of the cone dened by [y x[ = c[t t
0
[ with
the hyperplane dened by t = 0, see Figure 4.4
t
x
(x,t )
0
|yx|=c| tt |
0
Figure 4.4: Domain of dependence, case n = 3
4.2.2 Case n = 2
Consider the initial value problem
v
xx
+v
yy
= c
2
v
tt
(4.18)
v(x, y, 0) = f(x, y) (4.19)
v
t
(x, y, 0) = g(x, y), (4.20)
where f C
3
, g C
2
.
Using the formula for the solution of the three-dimensional initial value
problem we will derive a formula for the two-dimensional case. The following
consideration is called Hadamards method of decent.
Let v(x, y, t) be a solution of (4.18)-(4.20), then
u(x, y, z, t) := v(x, y, t)
is a solution of the three-dimensional initial value problem with initial data
f(x, y), g(x, y), independent of z, since u satises (4.18)-(4.20). Hence, since
u(x, y, z, t) = u(x, y, 0, t) +u
z
(x, y, z, t)z, 0 < < 1, and u
z
= 0, we have
v(x, y, t) = u(x, y, 0, t).
Poissons formula in the three-dimensional case implies
v(x, y, t) =
1
4c
2
t
_
1
t
_
B
ct
(x,y,0)
f(, ) dS
_
+
1
4c
2
t
_
B
ct
(x,y,0)
g(, ) dS. (4.21)
n
dS
S
_
+
S
r
d
Figure 4.5: Domains of integration
The integrands are independent on . The surface S is dened by
(, , ) := ( x)
2
+( y)
2
+
2
c
2
t
2
= 0. Then the exterior normal n
at S is n = /[[ and the surface element is given by dS = (1/[n
3
[)dd,
where the third coordinate of n is
n
3
=
_
c
2
t
2
( x)
2
( y)
2
ct
.
4.3. INHOMOGENEOUS EQUATION 117
The positive sign applies on S
+
, where > 0 and the sign is negative on S
where < 0, see Figure 4.5. We have S = S

+
S
.
Set =
_
( x)
2
+ ( y)
2
. Then it follows from (4.21)
Theorem 4.3. The solution of the Cauchy initial value problem (4.18)-
(4.20) is given by
v(x, y, t) =
1
2c
t
_
B
ct
(x,y)
f(, )
_
c
2
t
2
2
dd
+
1
2c
_
B
ct
(x,y)
g(, )
_
c
2
t
2
2
dd.
Corollary. In contrast to the three dimensional case, the domain of depen-
dence is here the disk B
ct
o
(x
0
, y
0
) and not the boundary only. Therefore, see
formula of Theorem 4.3, if f, g have supports in a compact domain D R
2
,
then these functions have inuence on the value v(x, y, t) for all time t > T,
T suciently large.
4.3 Inhomogeneous equation
Here we consider the initial value problem
2u = w(x, t) on x R
n
, t R (4.22)
u(x, 0) = f(x) (4.23)
u
t
(x, 0) = g(x), (4.24)
where 2u := u
tt
c
2
u. We assume f C
3
, g C
2
and w C
1
, which
are given.
Set u = u
1
+ u
2
, where u
1
is a solution of problem (4.22)-(4.24) with
w := 0 and u
2
is the solution where f = 0 and g = 0 in (4.22)-(4.24). Since
we have explicit solutions u
1
in the cases n = 1, n = 2 and n = 3, it remains
to solve
2u = w(x, t) on x R
n
, t R (4.25)
u(x, 0) = 0 (4.26)
u
t
(x, 0) = 0. (4.27)
The following method is called Duhamels principle which can be considered
as a generalization of the method of variations of constants in the theory of
ordinary dierential equations.
To solve this problem, we make the ansatz
u(x, t) =
_
t
0
v(x, t, s) ds, (4.28)
where v is a function satisfying
2v = 0 for all s (4.29)
and
v(x, s, s) = 0. (4.30)
From ansatz (4.28) and assumption (4.30) we get
u
t
= v(x, t, t) +
_
t
0
v
t
(x, t, s) ds,
=
_
t
0
v
t
(x, t, s). (4.31)
It follows u
t
(x, 0) = 0. The initial condition u(x, t) = 0 is satised because
of the ansatz (4.28). From (4.31) and ansatz (4.28) we see that
u
tt
= v
t
(x, t, t) +
_
t
0
v
tt
(x, t, s) ds,
x
u =
_
t
0
x
v(x, t, s) ds.
Therefore, since u is an ansatz for (4.25)-(4.27),
u
tt
c
2
x
u = v
t
(x, t, t) +
_
t
0
(2v)(x, t, s) ds
= w(x, t).
Thus necessarily v
t
(x, t, t) = w(x, t), see (4.29). We have seen that the
ansatz provides a solution of (4.25)-(4.27) if for all s
2v = 0, v(x, s, s) = 0, v
t
(x, s, s) = w(x, s). (4.32)
Let v
(x, t, s) be a solution of
2v = 0, v(x, 0, s) = 0, v
t
(x, 0, s) = w(x, s), (4.33)
then
v(x, t, s) := v
(x, t s, s)
is a solution of (4.32). In the case n = 3, where v
is given by, see Theorem

4.2,
v
(x, t, s) =
1
4c
2
t
_
B
ct
(x)
w(, s) dS
.
Then
v(x, t, s) = v
(x, t s, s)
=
1
4c
2
(t s)
_
B
c(ts)
(x)
w(, s) dS
.
from ansatz (4.28) it follows
u(x, t) =
_
t
0
v(x, t, s) ds
=
1
4c
2
_
t
0
_
B
c(ts)
(x)
w(, s)
t s
dS
ds.
Changing variables by = c(t s) yields
u(x, t) =
1
4c
2
_
ct
0
_
B
(x)
w(, t /c)
dS
d
=
1
4c
2
_
B
ct
(x)
w(, t r/c)
r
d,
where r = [x [.
Formulas for the cases n = 1 and n = 2 follow from formulas for the as-
sociated homogeneous equation with inhomogeneous initial values for these
cases.
Theorem 4.4. The solution of
2u = w(x, t), u(x, 0) = 0, u
t
(x, 0) = 0,
where w C
1
, is given by:
Case n = 3:
u(x, t) =
1
4c
2
_
B
ct
(x)
w(, t r/c)
r
d,
where r = [x [, x = (x
1
, x
2
, x
3
), = (
1
,
2
,
3
).
Case n = 2:
u(x, t) =
1
4c
_
t
0
_
_
B
c(t)
(x)
w(, )
_
c
2
(t )
2
r
2
d
_
d,
x = (x
1
, x
2
), = (
1
,
2
).
Case n = 1:
u(x, t) =
1
2c
_
t
0
_
_
x+c(t)
xc(t)
w(, ) d
_
d.
Remark. The integrand on the right in formula for n = 3 is called retarded
potential. The integrand is taken not at t, it is taken at an earlier time
t r/c.
4.4 A method of Riemann
Riemanns method provides a formula for the solution of the following
Cauchy initial value problem for a hyperbolic equation of second order in
two variables. Let
o : x = x(t), y = y(t), t
1
t t
2
,
be a regular curve in R
2
, that is, we assume x, y C
1
[t
1
, t
2
] and x
2
+y
2
,= 0.
Set
Lu := u
xy
+a(x, y)u
x
+b(x, y)u
y
+c(x, y)u,
where a, b C
1
and c, f C in a neighbourhood of o. Consider the initial
value problem
Lu = f(x, y) (4.34)
u
0
(t) = u(x(t), y(t)) (4.35)
p
0
(t) = u
x
(x(t), y(t)) (4.36)
q
0
(t) = u
y
(x(t), y(t)), (4.37)
where f C in a neighbourhood of o and u
0
, p
0
, q
0
C
1
are given.
We assume:
4.4. A METHOD OF RIEMANN 121
(i) u
0
(t) = p
0
(t)x
(t) +q
0
(t)y
(t) (strip condition),

(ii) o is not a characteristic curve. Moreover assume that the characteristic
curves, which are lines here and are dened by x = const. and y = const.,
have at most one point of intersection with o, and such a point is not a
touching point, i. e., tangents of the characteristic and o are dierent at
this point.
We recall that the characteristic equation to (4.34) is
x
y
= 0 which
is satised if
x
(x, y) = 0 or
y
(x, y) = 0. One family of characteris-
tics associated to these rst partial dierential of rst order is dened by
x
(t) = 1, y
(t) = 0, see Chapter 2.

Assume u, v C
1
and that u
xy
, v
xy
exist and are continuous. Dene the
adjoint dierential expression by
Mv = v
xy
(av)
x
(bv)
y
+cv.
We have
2(vLu uMv) = (u
x
v v
x
u + 2buv)
y
+ (u
y
v v
y
u + 2auv)
x
. (4.38)
Set
P = (u
x
v x
x
u + 2buv)
Q = u
y
v v
y
u + 2auv.
From (4.38) it follows for a domain R
2
2
_
(vLu uMv) dxdy =

_
(P
y
+Q
x
) dxdy
=
_
Pdx +Qdy, (4.39)
where integration in the line integral is anticlockwise. The previous equation
follows from Gauss theorem or after integration by parts:
_
(P
y
+Q
x
) dxdy =
_
(Pn
2
+Qn
1
) ds,
where n = (dy/ds, dx/ds), s arc length, (x(s), y(s)) represents .
Assume u is a solution of the initial value problem (4.34)-(4.37) and
suppose that v satises
Mv = 0 in .
A
B
S
x
y
0
P=(x ,y )
0
Figure 4.6: Riemanns method, domain of integration
Then, if we integrate over a domain as shown in Figure 4.6, it follows
from (4.39) that
2
_
vf dxdy =
_
BA
Pdx+Qdy+
_
AP
Pdx+Qdy+
_
PB
Pdx+Qdy. (4.40)
The line integral from B to A is known from initial data, see the denition
of P and Q.
Since
u
x
v v
x
u + 2buv = (uv)
x
+ 2u(bv v
x
),
it follows
_
AP
Pdx +Qdy =
_
AP
((uv)
x
+ 2u(bv v
x
)) dx
= (uv)(P) + (uv)(A)
_
AP
2u(bv v
x
) dx.
By the same reasoning we obtain for the third line integral
_
PB
Pdx +Qdy =
_
PB
((uv)
y
+ 2u(av v
y
)) dy
= (uv)(B) (uv)(P) +
_
PB
2u(av v
y
) dy.
4.4. A METHOD OF RIEMANN 123
Combining these equations with (4.39), we get
2v(P)u(P) =
_
BA
(u
x
v v
x
+ 2buv) dx (u
y
v v
y
u + 2auv) dy
+u(A)v(A) +u(B)v(B) + 2
_
AP
u(bv v
x
) dx
+2
_
PB
u(av v
y
) dy 2
_
fv dxdy. (4.41)
Let v be a solution of the initial value problem, see Figure 4.7 for the de-
nition of domain D(P),
x
y
0
P=(x ,y )
0
C
C
2
1
D(P)
Figure 4.7: Denition of Riemanns function
Mv = 0 in D(P) (4.42)
bv v
x
= 0 on C
1
(4.43)
av v
y
= 0 on C
2
(4.44)
v(P) = 1. (4.45)
Assume v satises (4.42)-(4.45), then
2u(P) = u(A)v(A) +u(B)v(B) 2
_
fv dxdy
=
_
BA
(u
x
v v
x
+ 2buv) dx (u
y
v v
y
u + 2auv) dy,
where the right hand side is known from given data.
A function v = v(x, y; x
0
, y
0
) satisfying (4.42)-(4.45) is called Riemanns
function.
Remark. Set w(x, y) = v(x, y; x
0
, y
0
) for xed x
0
, y
0
. Then (4.42)-(4.45)
imply
w(x, y
0
) = exp
__
x
x
0
b(, y
0
) d
_
on C
1
,
w(x
0
, y) = exp
__
y
y
0
a(x
0
, ) d
_
on C
2
.
Examples
1. u
xy
= f(x, y), then a Riemann function is v(x, y) 1.
2. Consider the telegraph equation of Chapter 3
u
tt
= c
2
x
u u
t
,
where u stands for one coordinate of electric or magnetic eld. Introducing
u = w(x, t)e
t
,
where = /(2), we arrive at
w
tt
=
c
2
x
w

2
4
2
.
Stretching the axis and transform the equation to the normal form we get
nally the following equation, the new function is denoted by u and the new
variables are denoted by x, y again,
u
xy
+cu = 0,
with a positive constant c. We make the ansatz for a Riemann function
v(x, y; x
0
, y
0
) = w(s), s = (x x
0
)(y y
0
)
and obtain
sw
+w
+cw = 0.
4.5. INITIAL-BOUNDARY VALUE PROBLEMS 125
Substitution =
4cs leads to Bessels dierential equation
2
z
() +z
() +
2
z() = 0,
where z() = w(
2
/(4c)). A solution is
J
0
() = J
0
_
_
4c(x x
0
)(y y
0
)
_
which denes a Riemann function since J
0
(0) = 1.
Remark. Bessels dierential equation is
x
2
y
(x) +xy
(x) + (x
2
n
2
)y(x) = 0,
where n R. If n N 0, then solutions are given by Bessel functions.
One of the two linearly independent solutions is bounded at 0. This bounded
solution is the Bessel function J
n
(x) of rst kind and of order n, see [1], for
example.
4.5 Initial-boundary value problems
In previous sections we looked at solutions dened for all x R
n
and t R.
In this and in the following section we seek solutions u(x, t) dened in a
bounded domain R
n
and for all t R and which satisfy additional
boundary conditions on .
4.5.1 Oscillation of a string
Let u(x, t), x [a, b], t R, be the deection of a string, see Figure 1.4 from
Chapter 1. Assume the deection occurs in the (x, u)-plane. This problem
is governed by the initial-boundary value problem
u
tt
(x, t) = u
xx
(x, t) on (0, l) (4.46)
u(x, 0) = f(x) (4.47)
u
t
(x, 0) = g(x) (4.48)
u(0, t) = u(l, t) = 0. (4.49)
Assume the initial data f, g are suciently regular. This implies compati-
bility conditions f(0) = f(l) = 0 and g(0) = g(l).
Fouriers method
To nd solutions of dierential equation (4.46) we make the separation of
variables ansatz
u(x, t) = v(x)w(t).
Inserting the ansatz into (4.46) we obtain
v(x)w
(t) = v
(x)w(t),
or, if v(x)w(t) ,= 0,
w
(t)
w(t)
=
v
(x)
v(x)
.
It follows, provided v(x)w(t) is a solution of dierential equation (4.46) and
v(x)w(t) ,= 0,
w
(t)
w(t)
= const. =:
and
v
(x)
v(x)
=
since x, t are independent variables.
Assume v(0) = v(l) = 0, then v(x)w(t) satises the boundary condi-
tion (4.49). Thus we look for solutions of the eigenvalue problem
v
(x) = v(x) in (0, l) (4.50)

v(0) = v(l) = 0, (4.51)
which has the eigenvalues
n
=
_
l
n
_
2
, n = 1, 2, . . . ,
and associated eigenfunctions are
v
n
= sin
_
l
nx
_
.
Solutions of
w
(t) =
n
w(t)
are
sin(
_
n
t), cos(
_
n
t).
Set
w
n
(t) =
n
cos(
_
n
t) +
n
sin(
_
n
t),
where
n
,
n
R. It is easily seen that w
n
(t)v
n
(x) is a solution of dierential
equation (4.46), and, since (4.46) is linear and homogeneous, also (principle
of superposition)
u
N
=
N
n=1
w
n
(t)v
n
(x)
which satises the dierential equation (4.46) and the boundary condi-
tions (4.49). Consider the formal solution of (4.46), (4.49)
u(x, t) =
n=1
_
n
cos(
_
n
t) +
n
sin(
_
n
t)
_
sin
_
_
n
x
_
. (4.52)
Formal means that we know here neither that the right hand side con-
verges nor that it is a solution of the initial-boundary value problem. For-
mally, the unknown coecients can be calculated from initial conditions (4.47),
(4.48) as follows. We have
u(x, 0) =
n=1
n
sin(
_
n
x) = f(x).
Multiplying this equation by sin(
k
x) and integrate over (0, l), we get
n
_
l
0
sin
2
(
_
k
x) dx =
_
l
0
f(x) sin(
_
k
x) dx.
We recall that
_
l
0
sin(
_
n
x) sin(
_
k
x) dx =
l
2
nk
.
Then
k
=
2
l
_
l
0
f(x) sin
_
k
l
x
_
dx. (4.53)
By the same argument it follows from
u
t
(x, 0) =
n=1
n
_
n
sin(
_
n
x) = g(x)
that
k
=
2
k
_
l
0
g(x) sin
_
k
l
x
_
dx. (4.54)
Under additional assumptions f C
4
0
(0, l), g C
3
0
(0, l) it follows that
the right hand side of (4.52), where
n
,
n
are given by (4.53) and (4.54),
respectively, denes a classical solution of (4.46)-(4.49) since under these
assumptions the series for u and the formal dierentiate series for u
t
, u
tt
,
u
x
, u
xx
converges uniformly on 0 x l, 0 t T, 0 < T < xed, see
an exercise.
4.5.2 Oscillation of a membrane
Let R
2
be a bounded domain. We consider the initial-boundary value
problem
u
tt
(x, t) =
x
u in R, (4.55)
u(x, 0) = f(x), x , (4.56)
u
t
(x, 0) = g(x), x , (4.57)
u(x, t) = 0 on R. (4.58)
As in the previous subsection for the string, we make the ansatz (separation
of variables)
u(x, t) = w(t)v(x)
which leads to the eigenvalue problem
v = v in , (4.59)
v = 0 on . (4.60)
Let
n
are the eigenvalues of (4.59), (4.60) and v
n
a complete associated
orthonormal system of eigenfunctions. We assume is suciently regular
such that the eigenvalues are countable, which is satised in the following
examples. Then the formal solution of the above initial-boundary value
problem is
u(x, t) =
n=1
_
n
cos(
_
n
t) +
n
sin(
_
n
t)
_
v
n
(x),
where
n
=
_
f(x)v
n
(x) dx
n
=
1
n
_
g(x)v
n
(x) dx.
Remark. In general, eigenvalues of (4.59), (4.59) are not known explicitly.
There are numerical methods to calculate these values. In some special
cases, see next examples, these values are known.
Examples
1. Rectangle membrane. Let
= (0, a) (0, b).
Using the method of separation of variables, we nd all eigenvalues of (4.59), (4.60)
which are given by
kl
=
_
k
2
a
2
+
l
2
b
2
, k, l = 1, 2, . . .
and associated eigenfunctions, not normalized, are
u
kl
(x) = sin
_
k
a
x
1
_
sin
_
l
b
x
2
_
.
2. Disk membrane. Set
= x R
2
: x
2
1
+x
2
2
< R
2
.
In polar coordinates, the eigenvalue problem (4.59), (4.60) is given by
1
r
_
(ru
r
)
r
+
1
r
u
_
= u (4.61)
u(R, ) = 0, (4.62)
here is u = u(r, ) := v(r cos , r sin ). We will nd eigenvalues and eigen-
functions by separation of variables
u(r, ) = v(r)q(),
where v(R) = 0 and q() is periodic with period 2 since u(r, ) is single
valued. This leads to
1
r
_
(rv
q +
1
r
vq
_
= vq.
Dividing by vq, provided vq ,= 0, we obtain
1
r
_
(rv
(r))
v(r)
+
1
r
q
()
q()
_
= , (4.63)
which implies
q
()
q()
= const. =: .
Thus, we arrive at the eigenvalue problem
q
() = q()
q() = q( + 2).
It follows that eigenvalues are real and nonnegative. All solutions of the
dierential equation are given by
q() = Asin(
) +Bcos(
),
where A, B are arbitrary real constants. From the periodicity requirement
Asin(
) +Bcos(
) = Asin(
( + 2)) +Bcos(
( + 2))
it follows
2
sin(
) (Acos(
) Bsin(
)) = 0,
which implies, since A, B are not zero simultaneously, because we are look-
ing for q not identically zero,
sin(
) sin(
+) = 0
for all and a = (A, B, ). Consequently the eigenvalues are
n
= n
2
, n = 0, 1, . . . .
Inserting q
()/q() = n
2
into (4.63), we obtain the boundary value prob-
lem
r
2
v
(r) +rv
(r) + (r
2
n
2
)v = 0 on (0, R) (4.64)
v(R) = 0 (4.65)
sup
r(0,R)
[v(r)[ < . (4.66)
Set z =
r and v(r) = v(z/
) =: y(z), then, see (4.64),

z
2
y
(z) +zy
(z) + (z
2
n
2
)y(z) = 0,
2
sin x sin y = 2 cos
x + y
2
sin
x y
2
cos x cos y = 2 sin
x + y
2
sin
x y
2
where z > 0. Solutions of this dierential equations which are bounded at
zero are Bessel functions of rst kind and n-th order J
n
(z). The eigenvalues
follows from boundary condition (4.65), i. e., from J
n
(
R) = 0. Denote
by
nk
the zeros of J
n
(z), then the eigenvalues of (4.61)-(4.61) are
nk
=
_
nk
R
_
2
and the associated eigenfunctions are
J
n
(
_
nk
r) sin(n), n = 1, 2, . . .
J
n
(
_
nk
r) cos(n), n = 0, 1, 2, . . . .
Thus the eigenvalues
0k
are simple and
nk
, n 1, are double eigenvalues.
Remark. For tables with zeros of J
n
(x) and for much more properties of
Bessel functions see [25]. One has, in particular, the asymptotic formula
J
n
(x) =
_
2
x
_
1/2
_
cos(x n/2 /5) +O
_
1
x
__
as x . It follows from this formula that there are innitely many zeros
of J
n
(x).
4.5.3 Inhomogeneous wave equations
Let R
n
be a bounded and suciently regular domain. In this section
we consider the initial-boundary value problem
u
tt
= Lu +f(x, t) in R (4.67)
u(x, 0) = (x) x (4.68)
u
t
(x, 0) = (x) x (4.69)
u(x, t) = 0 for x and t R
n
, (4.70)
where u = u(x, t), x = (x
1
, . . . , x
n
), f, , are given and L is an elliptic
dierential operator. Examples for L are:
1. L =
2
/x
2
, oscillating string.
2. L =
x
, oscillating membrane.
3.
Lu =
n
i,j=1
x
j
_
a
ij
(x)u
x
i
_
,
where a
ij
= a
ji
are given suciently regular functions dened on . We
assume L is uniformly elliptic, that is, there is a constant > 0 such that
n
i,j=1
a
ij
j
[[
2
for all x and R
n
.
4. Let u = (u
1
, . . . , u
m
) and
Lu =
n
i,j=1
x
j
_
A
ij
(x)u
x
i
_
,
where A
ij
= A
ji
are given suciently regular (m m)-matrices on . We
assume that L denes an elliptic system. An example for this case is the
linear elasticity.
Consider the eigenvalue problem
Lv = v in (4.71)
v = 0 on . (4.72)
Assume there are innitely many eigenvalues
0 <
1

2
. . .
and a system of associated eigenfunctions v
1
, v
2
, . . . which is complete and
orthonormal in L
2
(). This assumption is satised if is bounded and if
is suciently regular.
For the solution of (4.67)-(4.70) we make the ansatz
u(x, t) =
k=1
v
k
(x)w
k
(t), (4.73)
with functions w
k
(t) which will be determined later. It is assumed that all
series are convergent and that following calculations make sense. Let
f(x, t) =
k=1
c
k
(t)v
k
(x) (4.74)
be Fouriers decomposition of f with respect to the eigenfunctions v
k
. We
have
c
k
(t) =
_
f(x, t)v
k
(x) dx, (4.75)
which follows from (4.74) after multiplying with v
l
(x) and integrating over
.
Set
, v
k
) =
_
(x)v
k
(x) dx,
then
(x) =
k=1
, v
k
)v
k
(x)
(x) =
k=1
, v
k
)v
k
(x)
are Fouriers decomposition of and , respectively.
In the following we will determine w
k
(t), which occurs in ansatz (4.73),
from the requirement that u = v
k
(x)w
k
(t) is a solution of
u
tt
= Lu +c
k
(t)v
k
(x)
and that the initial conditions
w
k
(0) = , v
k
), w
k
(0) = , v
k
)
are satised. From the above dierential equation it follows
w
k
(t) =
k
w
k
(t) +c
k
(t).
Thus
w
k
(t) = a
k
cos(
_
k
t) +b
k
sin(
_
k
t) (4.76)
+
1
k
_
t
0
c
k
() sin(
_
k
(t )) d,
where
a
k
= , v
k
), b
k
=
1
k
, v
k
).
Summarizing, we have
Proposition 4.2. The (formal) solution of the initial-boundary value prob-
lem (4.67)-(4.70) is given by
u(x, t) =
k=1
v
k
(x)w
k
(t),
where v
k
is a complete orthonormal system of eigenfunctions of (4.71), (4.72)
and the functions w
k
are dened by (4.76).
The resonance phenomenon
Set in (4.67)-(4.70) = 0, = 0 and assume that the external force f is
periodic and is given by
f(x, t) = Asin(t)v
n
(x),
where A, are real constants and v
n
is one of the eigenfunctions of (4.71), (4.72).
It follows
c
k
(t) =
_
f(x, t)v
k
(x) dx = A
nk
sin(t).
Then the solution of the initial value problem (4.67)-(4.70) is
u(x, t) =
Av
n
(x)
n
_
t
0
sin() sin(
_
n
(t )) d
= Av
n
(x)
1
n
_

n
sin(
_
k
t) sin(t)
_
,
provided ,=
n
. It follows
u(x, t)
A
2
n
v
n
(x)
_
sin(
n
t)
n
t cos(
_
n
t)
_
if
n
. The right hand side is also the solution of the initial-boundary
value problem if =
n
.
Consequently [u[ can be arbitrarily large at some points x and at some
times t if =
n
. The frequencies
n
are called critical frequencies at
which resonance occurs.
A uniqueness result
The solution of of the initial-boundary value problem (4.67)-(4.70) is unique
in the class C
2
( R).
Proof. Let u
1
, u
2
are two solutions, then u = u
2
u
1
satises
u
tt
= Lu in R
u(x, 0) = 0 x
u
t
(x, 0) = 0 x
u(x, t) = 0 for x and t R
n
.
As an example we consider Example 3 from above and set
E(t) =
_
(
n
i,j=1
a
ij
(x)u
x
i
u
x
j
+u
t
u
t
) dx.
Then
E
(t) = 2
_
(
n
i,j=1
a
ij
(x)u
x
i
u
x
j
t
+u
t
u
tt
) dx
= 2
_
(
n
i,j=1
a
ij
(x)u
x
i
u
t
n
j
) dS
+2
_
u
t
(Lu +u
t
t) dx
= 0.
It follows E(t) = const. From u
t
(x, 0) = 0 and u(x, 0) = 0 we get E(0) = 0.
Consequently E(t) = 0 for all t, which implies, since L is elliptic, that
u(x, t) = const. on R. Finally, the homogeneous initial and boundary
value conditions lead to u(x, t) = 0 on R. 2
4.6 Exercises
1. Show that u(x, t) C
2
(R
2
) is a solution of the one-dimensional wave
equation
u
tt
= c
2
u
xx
if and only if
u(A) +u(C) = u(B) +u(D)
holds for all parallelograms ABCD in the (x, t)-plane, which are bounded
by characteristic lines, see Figure 4.8.
x
t
B
C
A
D
Figure 4.8: Figure to the exercise
2. Method of separation of variables: Let v
k
(x) be an eigenfunction to
the eigenvalue of the eigenvalue problem v
(x) = v(x) in (0, l),

v(0) = v(l) = 0 and let w
k
(t) be a solution of dierential equation
w
(t) =
k
w(t). Prove that v
k
(x)w
k
(t) is a solution of the partial
dierential equation (wave equation) u
tt
= u
xx
.
3. Solve for given f(x) and R the initial value problem
u
t
+u
x
+u
xxx
= 0 in R R
+
u(x, 0) = f(x) .
4. Let S := (x, t); t = x be spacelike, i. e., [[ < 1/c
2
) in (x, t)-space,
x = (x
1
, x
2
, x
3
). Show that the Cauchy initial value problem 2u = 0
4.6. EXERCISES 137
with data for u on S can be transformed using the Lorentz-transform
x
1
=
x
1
c
2
t
_
1
2
c
2
, x
2
= x
2
, x
3
= x
3
, t
=
t x
1
_
1
2
c
2
into the initial value problem, in new coordinates,
2u = 0
u(x
, 0) = f(x
)
u
t
(x
, 0) = g(x
) .
Here we denote the transformed function by u again.
5. (i) Show that
u(x, t) :=
n=1
n
cos
_
n
l
t
_
sin
_
n
l
x
_
is a C
2
-solution of the wave equation u
tt
= u
xx
if [
n
[ c/n
4
, where
the constant c is independent of n.
(ii) Set
n
:=
_
l
0
f(x) sin
_
n
l
x
_
dx.
Prove [
n
[ c/n
4
, provided f C
4
0
(0, l).
6. Let be the rectangle (0, a) (0, b). Find all eigenvalues and associ-
ated eigenfunctions of u = u in , u = 0 on .
Hint: Separation of variables.
7. Find a solution of Schrodingers equation
i
t
=

2
2m
x
+V (x) in R
n
R,
which satises the side condition
_
n
R
[(x, t)[
2
dx = 1 ,
provided E R is an (eigenvalue) of the elliptic equation
u +
2m
2
(E V (x))u = 0 in R
n
under the side condition
_
n
R
[u[
2
dx = 1, u : R
n
C.
Here is : R
n
R C, Plancks constant (a small positive con-
stant), V (x) a given potential.
Remark. In the case of a hydrogen atom the potential is V (x) =
e/[x[, e is here a positive constant. Then eigenvalues are given by
E
n
= me
4
/(2
2
n
2
), n N, see [22], pp. 202.
8. Find nonzero solutions by using separation of variables of u
tt
=
x
u
in (0, ), u(x, t) = 0 on , where is the circular cylinder
= (x
1
, x
2
, x
3
) R
n
: x
2
1
+x
2
2
< R
2
, 0 < x
3
< h.
3u
tt
4u
xx
= 0
u(x, 0) = sin x
u
t
(x, 0) = 1 .
u
tt
c
2
u
xx
= x
2
, t > 0, x R
u(x, 0) = x
u
t
(x, 0) = 0 .
Hint: Find a solution of the dierential equation independent on t,
and transform the above problem into an initial value problem with
homogeneous dierential equation by using this solution.
11. Find with the method of separation of variables nonzero solutions
u(x, t), 0 x 1, 0 t < , of
u
tt
u
xx
+u = 0 ,
such that u(0, t) = 0, and u(1, t) = 0 for all t [0, ).
12. Find solutions of the equation
u
tt
c
2
u
xx
=
2
u, = const.
which can be written as
u(x, t) = f(x
2
c
2
t
2
) = f(s), s := x
2
c
2
t
2
4.6. EXERCISES 139
with f(0) = K, K a constant.
Hint: Transform equation for f(s) by using the substitution s := z
2
/A
with an appropriate constant A into Bessels dierential equation
z
2
f
(z) +zf
(z) + (z
2
n
2
)f = 0, z > 0
with n = 0.
Remark. The above dierential equation for u is the transformed tele-
graph equation (see Section 4.4).
13. Find the formula for the solution of the following Cauchy initial value
problem u
xy
= f(x, y), where S: y = ax + b, a > 0, and the initial
conditions on S are given by
u = x +y +,
u
x
= ,
u
y
= ,
a, b, , , constants.
14. Find all eigenvalues of
q
() = q()
q() = q( + 2) .
Chapter 5
Fourier transform
Fouriers transform is an integral transform which can simplify investigations
for linear dierential or integral equations since it transforms a dierential
operator into an algebraic equation.
5.1 Denition, properties
Denition. Let f C
s
0
(R
n
), s = 0, 1, . . . . The function

f dened by
f() = (2)
n/2
_
R
n
e
ix
f(x) dx, (5.1)
where R
n
, is called Fourier transform of f, and the function g given by
g(x) = (2)
n/2
_
R
n
e
ix
g() d (5.2)
is called inverse Fourier transform, provided the integrals on the right hand
side exist.
From (5.1) it follows by integration by parts that dierentiation of a func-
tion is changed to multiplication of its Fourier transforms, or an analytical
operation is converted into an algebraic operation. More precisely, we have
Proposition 5.1.
f() = i
||
f(),
where [[ s.
141
142 CHAPTER 5. FOURIER TRANSFORM
The following proposition shows that the Fourier transform of f decreases
rapidly for [[ , provided f C
s
0
(R
n
). In particular, the right hand
side of (5.2) exists for g :=

f if f C
n+1
0
(R
n
).
Proposition 5.2. Assume g C
s
0
(R
n
), then there is a constant M =
M(n, s, g) such that
[ g()[
M
(1 +[[)
s
.
Proof. Let = (
1
, . . . ,
n
) be xed and let j be an index such that [
j
[ =
max
k
[
k
[. Then
[[ =
_
n
k=1
2
k
_
1/2
n[
j
[
which implies
(1 +[[)
s
=
s
k=0
_
s
k
_
[[
k
2
s
s
k=0
n
k/2
[
j
[
k
2
s
n
s/2
||s
[
[.
This inequality and Proposition 5.1 imply
(1 +[[)
s
[ g()[ 2
s
n
s/2
||s
[(i)
g()[
2
s
n
s/2
||s
_
R
n
[D
g(x)[ dx =: M.
2
The notation inverse Fourier transform for (5.2) is justied by
Theorem 5.1.

f = f and

f = f.
Proof. See [27], for example. We will prove the rst assertion
(2)
n/2
_
R
n
e
ix
f() d = f(x) (5.3)

5.1. DEFINITION, PROPERTIES 143
here. The proof of the other relation is left as an exercise. All integrals
appearing in the following exist, see Proposition 5.2 and the special choice
of g.
(i) Formula
_
R
n
g()
f()e
ix
d =
_
R
n
g(y)f(x +y) dy (5.4)
follows by direct calculation:
_
R
n
g()
_
(2)
n/2
_
R
n
e
ixy
f(y) dy
_
e
ix
d
= (2)
n/2
_
R
n
__
R
n
g()e
i(yx)
d
_
f(y) dy
=
_
R
n
g(y x)f(y) dy
=
_
R
n
g(y)f(x +y) dy.
(ii) Formula
(2)
n/2
_
R
n
e
iy
g() d =
n
g(y/) (5.5)
for each > 0 follows after substitution z = in the left hand side of (5.1).
(iii) Equation
_
R
n
g()
f()e
ix
d =
_
R
n
g(y)f(x +y) dy (5.6)
follows from (5.4) and (5.5). Set G() := g(), then (5.4) implies
_
R
n
G()
f()e
ix
d =
_
R
n
G(y)f(x +y) dy.

Since, see (5.5),
G(y) = (2)
n/2
_
R
n
e
iy
g() d
=
n
g(y/),
we arrive at
_
R
n
g()
f() =
_
R
n
n
g(y/)f(x +y) dy
=
_
R
n
g(z)f(x +z) dz.
Letting 0, we get
g(0)
_
R
n
f()e
ix
d = f(x)
_
R
n
g(y) dy. (5.7)
Set
g(x) := e
|x|
2
/2
,
then
_
R
n
g(y) dy = (2)
n/2
. (5.8)
Since g(0) = 1, the rst assertion of Theorem 5.1 follows from (5.7) and (5.8).
It remains to show (5.8).
(iv) Proof of (5.8). We will show
g(y) : = (2)
n/2
_
R
n
e
|x|
2
/2
e
ixx
dx
= e
|y|
2
/2
.
The proof of
_
R
n
e
|y|
2
/2
dy = (2)
n/2
is left as an exercise. Since
_
x
2
+i
y
2
_
_
x
2
+i
y
2
_
=
_
[x[
2
2
+ix y
[y[
2
2
_
it follows
_
R
n
e
|x|
2
/2
e
ixy
dx =
_
R
n
e
2
e
|y|
2
/2
dx
= e
|y|
2
/2
_
R
n
e
2
dx
= 2
n/2
e
|y|
2
/2
_
R
n
e
2
d
where
:=
x
2
+i
y
2
.
Consider rst the one-dimensional case. According to Cauchys theorem we
have
_
C
e
2
d = 0,
where the integration is along the curve C which is the union of four curves
as indicated in Figure 5.1.
Re
Im
C
C
C
C
i
y
2
4
3
1
2
R R
Figure 5.1: Proof of (5.8)
Consequently
_
C
3
e
2
d =
1
2
_
R
R
e
x
2
/2
dx
_
C
2
e
2
d
_
C
4
e
2
d.
It follows
lim
R
_
C
3
e
2
d =
since
lim
R
_
C
k
e
2
d = 0, k = 2, 4.
The case n > 1 can be reduced to the one-dimensional case as follows. Set
=
x
2
+i
y
2
= (
1
, . . . ,
n
),
where
l
=
x
l
2
+i
y
l
2
.
From d = d
1
. . . d
l
and
e
2
= e
n
l=1

2
l
=
n
l=1
e
2
l
it follows
_
R
n
e
2
d =
n
l=1
_
l
e
2
l
d
l
,
where for xed y
l
= z C : z =
x
l
2
+i
y
l
2
, < x
l
< +.
2
There is a useful class of functions for which the integrals in the denition
of

f and

f exist.
For u C
(R
n
) we set
q
j,k
(u) := max
: ||k
_
sup
R
n
_
(1 +[x[
2
)
j/2
[D
u(x)[
_
_
.
Denition. The Schwartz class of rapidly degreasing functions is
o(R
n
) = u C
(R
n
) : q
j,k
(u) < for any j, k N 0 .
This space is a Frechet space.
Proposition 5.3. Assume u o(R
n
), then u and u o(R
n
).
Proof. See [24], Chapter 1.2, for example, or an exercise.
5.1.1 Pseudodierential operators
The properties of Fourier transform lead to a general theory for linear partial
dierential or integral equations. In this subsection we dene
D
k
=
1
i
x
k
, k = 1, . . . , n,
and for each multi-index as in Subsection 3.5.1
D
= D
1
1
. . . D
n
n
.
Thus
D
=
1
i
||
||
x
1
1
. . . x
n
n
.
Let
p(x, D) :=
||m
a
(x)D
,
be a linear partial dierential of order m, where a
are given suciently

regular functions.
According to Theorem 5.1 and Proposition 5.3, we have, at least for
u o(R
n
),
u(x) = (2)
n/2
_
R
n
e
ix
u() d,
which implies
D
u(x) = (2)
n/2
_
R
n
e
ix
u() d.
Consequently
p(x, D)u(x) = (2)
n/2
_
R
n
e
ix
p(x, ) u() d, (5.9)
where
p(x, ) =
||m
a
(x)
.
The right hand side of (5.9) makes sense also for more general functions
p(x, ), not only for polynomials.
Denition. The function p(x, ) is called symbol and
(Pu)(x) := (2)
n/2
_
R
n
e
ix
p(x, ) u() d
is said to be pseudodierential operator.
An important class of symbols for which the right hand side in this deni-
tion of a pseudodierential operator is dened is S
m
which is the subset of
p(x, ) C
( R
n
) such that
[D
x
D
p(x, )[ C
K,,
(p) (1 +[[)
m||
for each compact K .
Above we have seen that linear dierential operators dene a class of pseu-
dodierential operators. Even integral operators can be written (formally)
as pseudodierential operators. Let
(Pu)(x) =
_
R
n
K(x, y)u(y) dy
be an integral operator. Then
(Pu)(x) = (2)
n/2
_
R
n
K(x, y)
_
R
n
e
ix
u() d
= (2)
n/2
_
R
n
e
ix
__
R
n
e
i(yx)
K(x, y) dy
_
u().
Then the symbol associated to the above integral operator is
p(x, ) =
_
R
n
e
i(yx)
K(x, y) dy.
5.2. EXERCISES 149
5.2 Exercises
1. Show
_
R
n
e
|y|
2
/2
dy = (2)
n/2
.
2. Show that u o(R
n
) implies u, u o(R
n
).
3. Give examples for functions p(x, ) which satisfy p(x, ) S
m
.
4. Find a formal solution of Cauchys initial value problem for the wave
equation by using Fouriers transform.
Chapter 6
Parabolic equations
Here we consider linear parabolic equations of second order. An example is
the heat equation
u
t
= a
2
u,
where u = u(x, t), x R
3
, t 0, and a
2
is a positive constant called
conductivity coecient. The heat equation has its origin in physics where
u(x, t) is the temperature at x at time t, see [20], p. 394, for example.
Remark 1. After scaling of axis we can assume a = 1.
Remark 2. By setting t := t, the heat equation changes to an equation
which is called backward equation. This is the reason for the fact that
the heat equation describes irreversible processes in contrast to the wave
equation 2u = 0 which is invariant with respect the mapping t t.
Mathematically, it means that it is not possible, in general, to nd the
distribution of temperature at an earlier time t < t
0
if the distribution is
given at t
0
.
Consider the initial value problem for u = u(x, t), u C
(R
n
R
+
),
u
t
= u in x R
n
, t 0, (6.1)
u(x, 0) = (x), (6.2)
where C(R
n
) is given and
x
.
151
152 CHAPTER 6. PARABOLIC EQUATIONS
6.1 Poissons formula
Assume u is a solution of (6.1), then, since Fourier transform is a linear
mapping,
u
t
u =

0.
From properties of the Fourier transform, see Proposition 5.1, we have
u =
n
k=1
2
u
x
2
k
=
n
k=1
i
2
2
k
u(),
provided the transforms exist. Thus we arrive at the ordinary dierential
equation for the Fourier transform of u
d u
dt
+[[
2
u = 0,
where is considered as a parameter. The solution is
u(, t) =

()e
||
2
t
since u(, 0) =

(). From Theorem 5.1 it follows
u(x, t) = (2)
n/2
_
R
n
()e
||
2
t
e
ix
d
= (2)
n
_
R
n
(y)
__
R
n
e
i(xy)||
2
t
d
_
dy.
Set
K(x, y, t) = (2)
n
_
R
n
e
i(xy)||
2
t
d.
By the same calculations as in the proof of Theorem 5.1, step (vi), we nd
K(x, y, t) = (4t)
n/2
e
|xy|
2
/4t
. (6.3)
Thus we have
u(x, t) =
1
_
2
t
_
n
_
R
n
(z)e
|xz|
2
/4t
dz. (6.4)
Denition. Formula (6.4) is called Poissons formula and the function K
dened by (6.3) heat kernel or fundamental solution of the heat equation.
6.1. POISSONS FORMULA 153
K
K(x,y,t )
K(x,y,t )
1
2
Figure 6.1: Kernel K(x, y, t), = [x y[, t
1
< t
2
Proposition 6.1 The kernel K has following properties:
(i) K(x, y, t) C
(R
n
R
n
R
+
),
(ii) (/t )K(x, y, t) = 0, t > 0,
(iii) K(x, y, t) > 0, t > 0,
(iv)
_
R
n
K(x, y, t) dy = 1, x R
n
, t > 0,
(v) For each xed > 0:
lim
t 0
t > 0
_
R
n
\B
(x)
K(x, y, t) dy = 0
uniformly for x R
n
.
Proof. (i) and (iii) are obviously, and (ii) follows from the denition of K.
Equations (iv) and (v) hold since
_
R
n
\B
(x)
K(x, y, t) dy =
_
R
n
\B
(x)
(4t)
n/2
e
|xy|
2
/4t
dy
=
n/2
_
R
n
\B
/
4t
(0)
e
||
2
d
by using the substitution y = x+(4t)
1/2
. For xed > 0 it follows (v) and
for := 0 we obtain (iv). 2
Theorem 6.1. Assume C(R
n
) and sup
R
n [(x)[ < . Then u(x, t)
given by Poissons formula (6.4) is in C
(R
n
R
+
), continuous on R
n
[0, ) and a solution of the initial value problem (6.1), (6.2).

Proof. It remains to show
lim
x
t 0
u(x, t) = ().
Since is continuous there exists for given > 0 a = () such that [(y)
x
t
+
+2
Figure 6.2: Figure to the proof of Theorem 6.1
()[ < if [y [ < 2. Set M := sup
R
n [(y)[. Then, see Proposition 6.1,
u(x, t) () =
_
R
n
K(x, y, t) ((y) ()) dy.
6.2. INHOMOGENEOUS HEAT EQUATION 155
It follows, if [x [ < and t > 0, that
[u(x, t) ()[
_
B
(x)
K(x, y, t) [(y) ()[ dy
+
_
R
n
\B
(x)
K(x, y, t) [(y) ()[ dy
_
B
2
(x)
K(x, y, t) [(y) ()[ dy
+2M
_
R
n
\B
(x)
K(x, y, t) dy

_
R
n
K(x, y, t) dy + 2M
_
R
n
\B
(x)
K(x, y, t) dy
< 2
if 0 < t t
0
, t
0
suciently small. 2
Remarks. 1. Uniqueness follows under the additional growth assumption
[u(x, t)[ Me
a|x|
2
in D
T
,
where M and a are positive constants, see Proposition 6.2 below.
In the one-dimensional case, one has uniqueness in the class u(x, t) 0 in
D
T
, see [10], pp. 222.
2. u(x, t) dened by Poissons formula depends on all values (y), y R
n
.
That means, a perturbation of , even far from a xed x, has inuence to the
value u(x, t). In physical terms, this means that heat travels with innite
speed, in contrast to the experience.
6.2 Inhomogeneous heat equation
Here we consider the initial value problem for u = u(x, t), u C
(R
n
R
+
),
u
t
u = f(x, t) in x R
n
, t 0,
u(x, 0) = (x),
where and f are given. From
u
t
u =

f(x, t)
we obtain an initial value problem for an ordinary dierential equation:
d u
dt
+[[
2
u =

f(, t)
u(, 0) =

().
The solution is given by
u(, t) = e
||
2
t
() +
_
t
0
e
||
2
(t)
f(, ) d.
Applying the inverse Fourier transform and a calculation as in the proof of
Theorem 5.1, step (vi), we get
u(x, t) = (2)
n/2
_
R
n
e
ix
_
e
||
2
t
()
+
_
t
0
e
||
2
(t)
f(, ) d
_
d.
From the above calculation for the homogeneous problem and calculation as
in the proof of Theorem 5.1, step (vi), we obtain the formula
u(x, t) =
1
(2
t)
n
_
R
n
(y)e
|yx|
2
/(4t)
dy
+
_
t
0
_
R
n
f(y, )
1
_
2
_
(t )
_
n
e
|yx|
2
/(4(t))
dy d.
This function u(x, t) is a solution of the above inhomogeneous initial value
problem provided
C(R
n
), sup
R
n
[(x)[ <
and if
f C(R
n
[0, )), M() := sup
R
n
[f(y, )[ < , 0 < .
6.3 Maximum principle
Let R
n
be a bounded domain. Set
D
T
= (0, T), T > 0,
S
T
= (x, t) : (x, t) 0 or (x, t) [0, T],
6.3. MAXIMUM PRINCIPLE 157
T
D
S
T
x
t
T

T
Figure 6.3: Notations to the maximum principle
see Figure 6.3
Theorem 6.2. Assume u C(D
T
), that u
t
, u
x
i
x
k
exist and are continuous
in D
T
, and
u
t
u 0 in D
T
.
Then
max
D
T
u(x, t) = max
S
T
u.
Proof. Assume initially u
t
u < 0 in D
T
. Let > 0 be small and
0 < < T. Since u C(D
T
), there is an (x
0
, t
0
) D
T
such that
u(x
0
, t
0
) = max
D
T
u(x, t).
Case (i). Let (x
0
, t
0
) D
T
. Hence, since D
T
is open, u
t
(x
0
, t
0
) = 0,
u
x
l
(x
0
, t
0
) = 0, l = 1, . . . , n and
n
l,k=1
u
x
l
x
k
(x
0
, t
0
)
l
k
0 for all R
n
.
The previous inequality implies that u
x
k
x
k
(x
0
, t
0
) 0 for each k. Thus we
arrived at a contradiction to u
t
u < 0 in D
T
.
Case (ii). Assume (x
0
, t
0
) T . Then it follows as above u 0 in
(x
0
, t
0
), and from u(x
0
, t
0
) u(x
0
, t), t t
0
, one concludes that u
t
(x
0
, t
0
)
0. We arrived at a contradiction to u
t
u < 0 in D
T
again.
Summarizing, we have shown that
max
D
T
u(x, t) = max
T
u(x, t).
Thus there is an (x
, t
) S
T
such that
u(x
, t
) = max
D
T
u(x, t).
Since u is continuous on D
T
, we have
lim
0
max
D
T
u(x, t) = max
D
T
u(x, t).
It follows that there is (x, t) S
T
such that
u(x, t) = max
D
T
u(x, t)
since S
T
S
T
and S
T
is compact. Thus, theorem is shown under the
assumption u
t
u < 0 in D
T
. Now assume u
t
u 0 in D
T
. Set
v(x, t) := u(x, t) kt,
where k is a positive constant. Then
v
t
v = u
t
u k < 0.
From above we have
max
D
T
u(x, t) = max
D
T
(v(x, t) +kt)
max
D
T
v(x, t) +kT
= max
S
T
v(x, t) +kT
max
S
T
u(x, t) +kT,
Letting k 0, we obtain
max
D
T
u(x, t) max
S
T
u(x, t).
Since S
T
D
T
, the theorem is shown. 2
If we replace in the above theorem the bounded domain by R
n
, then the
result remains true provided we assume an additional growth assumption
for u. More precisely, we have the following result which is a corollary of
the previous theorem. Set for a xed T, 0 < T < ,
D
T
= (x, t) : x R
n
, 0 < t < T.
Proposition 6.2. Assume u C(D
T
), that u
t
, u
x
i
x
k
exist and are contin-
uous in D
T
,
u
t
u 0 in D
T
,
and additionally that u satises the growth condition
u(x, t) Me
a|x|
2
,
where M and a are positive constants. Then
max
D
T
u(x, t) = max
S
T
u.
It follows immediately the
Corollary. The initial value problem u
t
u = 0 in D
T
, u(x, 0) = f(x),
x R
n
, has a unique solution in the class dened by u C(D
T
), u
t
, u
x
i
x
k
exist and are continuous in D
T
and [u(x, t)[ Me
a|x|
2
.
Proof of Proposition 6.2. See [10], pp. 217. We can assume that 4aT < 1,
since the nite interval can be divided into nite many intervals of equal
length with 4a < 1. Then we conclude successively for k that
u(x, t) sup
yR
n
u(y, k) sup
yR
n
u(y, 0)
for k t (k + 1), k = 0, . . . , N 1, where N = T/.
There is an > 0 such that 4a(T + ) < 1. Consider the comparison
function
v
(x, t) : = u(x, t) (4(T + t))

n/2
e
|xy|
2
/(4(T+t))
= u(x, t) K(ix, iy, T + t)
for xed y R
n
and for a constant > 0. Since the heat kernel K(ix, iy, t)
satises K
t
= K
x
, we obtain
t
v
= u
t
u 0.
Set for a constant > 0
D
T,
= (x, t) : [x y[ < , 0 < t < T.
Then we obtain from Theorem 6.2 that
v
(y, t) max
S
T,
v
,
where S
T,
S
T
of Theorem 6.2 with = B
(y), see Figure 6.3. On the

bottom of S
T,
we have, since K > 0,
v
(x, 0) u(x, 0) sup

zR
n
f(z).
On the cylinder part [x y[ = , 0 t T, of S
T,
it is
v
(x, t) Me
a|x|
2
(4(T + t))
n/2
e
2
/(4(T+t))
Me
a(|y|+)
2
(4(T +))
n/2
e
2
/(4(T+))
sup
zR
n
f(z)
for all >
0
(),
0
suciently large. We recall that 4a(T + ) < 1.
Summarizing, we have
max
S
T,
v
(x, t) sup
zR
n
f(z)
if >
0
(). Thus
v
(y, t) max
S
T,
v
(x, t) sup
zR
n
f(z)
if >
0
(). Since
v
(y, t) = u(y, t) (4(T + t))

n/2
it follows
u(y, t) (4(T + t))
n/2
sup
zR
n
f(z).
Letting 0, we obtain the assertion of the proposition. 2
The above maximum principle of Theorem 6.2 holds for a large class of
parabolic dierential operators, even for degenerate equations. Set
Lu =
n
i,j=1
a
ij
(x, t)u
x
i
x
j
,
where a
ij
C(D
T
) are real, a
ij
= a
ji
, and the matrix (a
ij
) is nonnegative,
that is,
n
i,j=1
a
ij
(x, t)
i
j
0 for all R
n
,
and (x, t) D
T
.
Theorem 6.3. Assume u C(D
T
), that u
t
, u
x
i
x
k
exist and are continuous
in D
T
, and
u
t
Lu 0 in D
T
.
Then
max
D
T
u(x, t) = max
S
T
u.
Proof. (i) One proof is a consequence of the following lemma: Let A, B real,
symmetric and nonnegative matrices. Nonnegative means that all eigenval-
ues are nonnegative. Then trace (AB)
n
i,j=1
a
ij
b
ij
0, see an exercise.
(ii) Another proof exploits transform to principle axis directly: Let U =
(z
1
, . . . , z
n
), where z
l
is an orthonormal system of eigenvectors to the eigen-
values
l
of the matrix A = (a
i,j
(x
0
, t
0
)). Set = U, x = U
T
(x x
0
)y and
v(y) = u(x
0
+Uy, t
0
), then
0
n
i,j=1
a
ij
(x
0
, t
0
)
i
j
=
n
i=1
2
i
0
n
i,j=1
u
x
i
x
j
j
=
n
i=1
v
y
i
y
i
2
i
.
It follows
i
0 and v
y
i
y
i
0 for all i. Consequently
n
i,j=1
a
ij
(x
0
, t
0
)u
x
i
x
j
(x
0
, t
0
) =
n
i=1
i
v
y
i
y
i
0.
2
6.4 Initial-boundary value problem
Consider the initial-boundary value problem for c = c(x, t)
c
t
= Dc in (0, ) (6.5)
c(x, 0) = c
0
(x) x (6.6)
c
n
= 0 on (0, ). (6.7)
Here is R
n
, n the exterior unit normal at the smooth parts of , D a
positive constant and c
0
(x) a given function.
Remark. In application to diusion problems, c(x, t) is the concentration
of a substance in a solution, c
0
(x) its initial concentration and D the coef-
cient of diusion.
First Ficks rule says that w = Dc/n, where w is the ow of the substance
through the boundary . Thus according to the Neumann boundary con-
dition (6.7), we assume that there is no ow through the boundary.
6.4.1 Fouriers method
Separation of variables ansatz c(x, t) = v(x)w(t) leads to the eigenvalue
problem, see the arguments of Section 4.5,
v = v in (6.8)
v
n
= 0 on , (6.9)
and to the ordinary dierential equation
w
(t) +Dw(t) = 0. (6.10)

Assume is bounded and suciently regular, then the eigenvalues
of (6.8), (6.9) are countable and
0 =
0
<
1

2
. . . .
Let v
j
(x) be a complete system of orthonormal (in L
2
()) eigenfunctions.
Solutions of (6.10) are
w
j
(t) = C
j
e
D
j
t
,
where C
j
are arbitrary constants.
6.4. INITIAL-BOUNDARY VALUE PROBLEM 163
According to the superposition principle,
c
N
(x, t) :=
N
j=0
C
j
e
D
j
t
v
j
(x)
is a solution of the dierential equation (6.8) and
c(x, t) :=
j=0
C
j
e
D
j
t
v
j
(x),
with
C
j
=
_
c
0
(x)v
j
(x) dx,
is a formal solution of the initial-boundary value problem (6.5)-(6.7).
Diusion in a tube
Consider a solution in a tube, see Figure 6.4. Assume the initial concentra-
l
Q
x
x
x
3
1
2
Figure 6.4: Diusion in a tube
tion c
0
(x
1
, x
2
, x
3
) of the substrate in a solution is constant if x
3
= const.
It follows from a uniqueness result below that the solution of the initial-
boundary value problem c(x
1
, x
2
, x
3
, t) is independent of x
1
and x
2
.
Set z = x
3
, then the above initial-boundary value problem reduces to
c
t
= Dc
zz
c(z, 0) = c
0
(z)
c
z
= 0, z = 0, z = l.
The (formal) solution is
c(z, t) =
n=0
C
n
e
D(
l
n)
2
t
cos
_
l
nz
_
,
where
C
0
=
1
l
_
l
0
c
0
(z) dz
C
n
=
2
l
_
l
0
c
0
(z) cos
_
l
nz
_
dz, n 1.
6.4.2 Uniqueness
Suciently regular solutions of the initial-boundary value problem (6.5)-
(6.7) are uniquely determined since from
c
t
= Dc in (0, )
c(x, 0) = 0
c
n
= 0 on (0, ).
it follows that for each > 0
0 =
_

0
_
(c
t
c D(c)c) dxdt
=
_
_

0
1
2
t
(c
2
) dtdx +D
_
_

0
[
x
c[
2
dxdt
=
1
2
_
c
2
(x, ) dx +D
_
_

0
[
x
c[
2
dxdt.
6.5 Black-Scholes equation
Solutions of the Black-Scholes equation dene the value of a derivative, for
example of a call or put option, which is based on an asset. An asset
6.5. BLACK-SCHOLES EQUATION 165
can be a stock or a derivative again, for example. In principle, there are
innitely many such products, for example n-th derivatives. The Black-
Scholes equation for the value V (S, t) of a derivative is
V
t
+
1
2
2
S
2
V
SS
+rSV
S
rV = 0 in , (6.11)
where for a xed T, 0 < T < ,
= (S, t) R
2
: 0 < S < , 0 < t < T,
and , r are positive constants. More precisely,
is the volatility of the underlying asset S,
r is the guaranteed interest rate of a risk-free investment.
If S(t) is the value of an asset at time t, then V (S(t), t) is the value
of the derivative at time t, where V (S, t) is the solution of an appropriate
initial-boundary value problem for the Black-Scholes equation, see below.
The Black-Scholes equation follows from Itos Lemma under some as-
sumptions on the random function associated to S(t), see [26], for example.
Call option
Here is V (S, t) := C(S, t), where C(S, t) is the value of the (European) call
option. In this case we have following side conditions to (6.11):
C(S, T) = maxS E, 0 (6.12)
C(0, t) = 0 (6.13)
C(S, t) = S +o(S) as S , uniformly in t, (6.14)
where E and T are positive constants, E is the exercise price and T the
expiry.
Side condition (6.12) means that the value of the option has no value at
time T if S(T) E,
condition (6.13) says that it makes no sense to buy assets if the value of
the asset is zero,
condition (6.14) means that we buy assets if its value becomes large, see
Figure 6.5, where the side conditions are indicated.
Theorem 6.4 (Black-Scholes formula for European call options). The so-
lution C(S, t), 0 S < , 0 t T, of the initial-boundary value prob-
lem (6.11)-(6.14) is explicitly known and is given by
C(S, t) = SN(d
1
) Ee
r(Tt)
N(d
2
),
S
t
T
C=max{SE,0}
C=0
C~S
Figure 6.5: Side conditions for a call option
where
N(x) =
1
2
_
x
e
y
2
/2
dy,
d
1
=
ln(S/E) + (r +
2
/2)(T t)
T t
,
d
2
=
ln(S/E) + (r
2
/2)(T t)
T t
.
Proof. Substitutions
S = Ee
x
, t = T

2
/2
, C = Ev(x, )
change equation (6.11) to
v
= v
xx
+ (k 1)v
x
kv, (6.15)
where
k =
r
2
/2
.
Initial condition (6.15) implies
v(x, 0) = maxe
x
1, 0. (6.16)
For a solution of (6.15) we make the ansatz
v = e
x+
u(x, ),
where and are constants which we will determine as follows. Inserting
the ansatz into dierential equation (6.15), we get
u +u
=
2
u + 2u
x
+u
xx
+ (k 1)(u +u
x
) ku.
Set =
2
+ (k 1) k and choose such that 0 = 2 + (k 1), then
u
= u
xx
. Thus
v(x, ) = e
(k1)x/2(k+1)
2
/4
u(x, ), (6.17)
where u(x, ) is a solution of the initial value problem
u
= u
xx
, < x < , > 0
u(x, 0) = u
0
(x),
with
u
0
(x) = max
_
e
(k+1)x/2
e
(k1)x/2
, 0
_
.
A solution of this initial value problem is given by Poissons formula
u(x, ) =
1
2
_
+
u
0
(s)e
(xs)
2
/(4)
ds.
Changing variable by q = (s x)/(
2), we get
u(x, ) =
1
2
_
+
u
0
(q
2 +x)e
q
2
/2
dq
= I
1
I
2
,
where
I
1
=
1
2
_

x/(
2)
e
(k+1)(x+q
2)
e
q
2
/2
dq
I
2
=
1
2
_

x/(
2)
e
(k1)(x+q
2)
e
q
2
/2
dq.
An elementary calculation shows that
I
1
= e
(k+1)x/2+(k+1)
2
/4
N(d
1
)
I
2
= e
(k1)x/2+(k1)
2
/4
N(d
2
),
where
d
1
=
x
2
+
1
2
(k + 1)
2
d
2
=
x
2
+
1
2
(k 1)
2
N(d
i
) =
1
2
_
d
i
e
s
2
/2
ds, i = 1, 2.
Combining the formula for u(x, ), denition (6.17) of v(x, ) and the previ-
ous settings x = ln(S/E), =
2
(T t)/2 and C = Ev(x, ), we get nally
the formula of of Theorem 6.4.
In general, the solution u of the initial value problem for the heat equa-
tion is not uniquely dened, see for example [10], pp. 206.
Uniqueness. The uniqueness follows from the growth assumption (6.14).
Assume there are two solutions of (6.11), (6.12)-(6.14), then the dierence
W(S, t) satises the dierential equation (6.11) and the side conditions
W(S, T) = 0, W(0, t) = 0, W(S, t) = O(S) as S
uniformly in 0 t T.
From a maximum principle consideration, see an exercise, it follows that
[W(S, t)[ cS on S 0, 0 t T. The constant c is independent on S
and t. From the denition of u we see that
u(x, ) =
1
E
e
x
W(S, t),
where S = Ee
x
, t = T 2/(
2
). Thus we have the growth property
[u(x, )[ Me
a|x|
, x R, (6.18)
with positive constants M and a. Then the solution of u
= u
xx
, in <
x < , 0
2
T/2, with the initial condition u(x, 0) = 0 is uniquely
dened in the class of functions satisfying the growth condition (6.18), see
Proposition 6.2 of this chapter. That is, u(x, ) 0. 2
Put option
Here is V (S, t) := P(S, t), where P(S, t) is the value of the (European) put
option. In this case we have following side conditions to (6.11):
P(S, T) = maxE S, 0 (6.19)
P(0, t) = Ee
r(Tt)
(6.20)
P(S, t) = o(S) as S , uniformly in 0 t T. (6.21)
Here E is the exercise price and T the expiry.
Side condition (6.19) means that the value of the option has no value at
time T if S(T) E,
condition (6.20) says that it makes no sense to sell assets if the value of the
asset is zero,
condition (6.21) means that it makes no sense to sell assets if its value
becomes large.
Theorem 6.5 (Black-Scholes formula for European put options). The solu-
tion P(S, t), 0 < S < , t < T of the initial-boundary value problem (6.11),
(6.19)-(6.21) is explicitly known and is given by
P(S, t) = Ee
r(Tt)
N(d
2
) SN(d
1
)
where N(x), d
1
, d
2
are the same as in Theorem 6.4.
Proof. The formula for the put option follows by the same calculations as
in the case of a call option or from the put-call parity
C(S, t) P(S, t) = S Ee
r(Tt)
and from
N(x) +N(x) = 1.
Concerning the put-call parity see an exercise. See also [26], pp. 40, for a
heuristic argument which leads to the formula for the put-call parity. 2
6.6 Exercises
1. Show that the solution u(x, t) given by Poissons formula satises
inf
zR
n
(z) u(x, t) sup
zR
n
(z) ,
provided (x) is continuous and bounded on R
n
.
2. Solve for given f(x) and R the initial value problem
u
t
+u
x
+u
xxx
= 0 in R R
+
u(x, 0) = f(x) .
3. Show by using Poissons formula:
(i) Each function f C([a, b]) can be approximated uniformly by a
sequence f
n
C
[a, b] .
(ii) In (i) we can choose polynomials f
n
(Weierstrasss approximation
theorem).
Hint: Concerning (ii), replace the kernel K = exp(
|yx|
2
4t
) by a se-
quence of Taylor polynomials in the variable z =
|yx|
2
4t
.
4. Let u(x, t) be a positive solution of
u
t
= u
xx
, t > 0,
where is a constant. Show that := 2u
x
/u is a solution of
Burgers equation
t
+
x
=
xx
, t > 0.
5. Assume u
1
(s, t), ..., u
n
(s, t) are solutions of u
t
= u
ss
. Show that
n
k=1
u
k
(x
k
, t)
is a solution of the heat equation u
t
u = 0 in R
n
(0, ).
6. Let A, B are real, symmetric and nonnegative matrices. Nonnegative
means that all eigenvalues are nonnegative. Prove that trace (AB)
n
i,j=1
a
ij
b
ij
0.
Hint: (i) Let U = (z
1
, . . . , z
n
), where z
l
is an orthonormal system of
eigenvectors to the eigenvalues
l
of the matrix B. Then
X = U
_
_
_
_
1
0 0
0
2
0
. . . . . . . . . . . . . . . . . . . . . .
0 0
n
_
_
_
_
U
T
6.6. EXERCISES 171
is a square root of B. We recall that
U
T
BU =
_
_
_
_
1
0 0
0
2
0
. . . . . . . . . . . . . . . .
0 0
n
_
_
_
_
.
(ii) trace (QR) =trace (RQ).
(iii) Let
1
(C), . . .
n
(C) are the eigenvalues of a real symmetric nn-
matrix. Then trace C =
n
l=1
l
(C), which follows from the funda-
mental lemma of algebra:
det (I C) =
n
(c
11
+. . . +c
nn
)
n1
+. . .
(
1
) . . . (
n
)
=
n
(
1
+. . . +
n
)
n+1
+. . .
7. Assume is bounded, u is a solution of the heat equation and u sat-
ises the regularity assumptions of the maximum principle (Theorem
6.2). Show that u achieves its maximum and its minimum on S
T
.
8. Prove the following comparison principle: Assume is bounded and
u, v satisfy the regularity assumptions of the maximum principle. Then
u
t
u v
t
v in D
T
u v on S
T
imply that u v in D
T
.
9. Show that the comparison principle implies the maximum principle.
10. Consider the boundary-initial value problem
u
t
u = f(x, t) in D
T
u(x, t) = (x, t) on S
T
,
where f, are given.
Prove uniqueness in the class u, u
t
, u
x
i
x
k
C(D
T
).
11. Assume u, v
1
, v
2
C
2
(D
T
) C(D
T
), and u is a solution of the
previous boundary-initial value problem and v
1
, v
2
satisfy
(v
1
)
t
v
1
f(x, t) (v
2
)
t
v
2
in D
T
v
1
v
2
on S
T
.
Show that (inclusion theorem)
v
1
(x, t) u(x, t) v
2
(x, t) on D
T
.
12. Show by using the comparison principle: let u be a suciently regular
solution of
u
t
u = 1 in D
T
u = 0 on S
T
,
then 0 u(x, t) t in D
T
.
13. Discuss the result of Theorem 6.3 for the case
Lu =
n
i,j=1
a
ij
(x, t)u
x
i
x
j
+
n
i
b
i
(x, t)u
x
i
+c(x, t)u(x, t).
14. Show that
u(x, t) =
n=1
c
n
e
n
2
t
sin(nx),
where
c
n
=
2
_

0
f(x) sin(nx) dx,
is a solution of the initial-boundary value problem
u
t
= u
xx
, x (0, ), t > 0,
u(x, 0) = f(x),
u(0, t) = 0,
u(, t) = 0,
if f C
4
(R) is odd with respect to 0 and 2-periodic.
15. (i) Find the solution of the diusion problem c
t
= Dc
zz
in 0 z l,
0 t < , D = const. > 0, under the boundary conditions c
z
(z, t) =
0 if z = 0 and z = l and with the given initial concentration
c(z, 0) = c
0
(z) :=
_
c
0
= const. if 0 z h
0 if h < z l.
(ii) Calculate lim
t
c(z, t).
6.6. EXERCISES 173
16. Prove the Black-Scholes Formel for an European put option.
Hint: Put-call parity.
17. Prove the put-call parity for European options
C(S, t) P(S, t) = S Ee
r(Tt)
by using the following uniqueness result: Assume W is a solution
of (6.11) under the side conditions W(S, T) = 0, W(0, t) = 0 and
W(S, t) = O(S) as S , uniformly on 0 t T. Then W(S, t)
0.
18. Prove that a solution V (S, t) of the initial-boundary value problem (6.11)
in under the side conditions (i) V (S, T) = 0, S 0, (ii) V (0, t) = 0,
0 t T, (iii) lim
S
V (S, t) = 0 uniformly in 0 t T, is uniquely
determined in the class C
2
() C().
19. Prove that a solution V (S, t) of the initial-boundary value problem (6.11)
in , under the side conditions (i) V (S, T) = 0, S 0, (ii) V (0, t) = 0,
0 t T, (iii) V (S, t) = S +o(S) as S , uniformly on 0 t T,
satises [V (S, t)[ cS for all S 0 and 0 t T.
Chapter 7
Elliptic equations of second
order
Here we consider linear elliptic equations of second order, mainly the Laplace
equation
u = 0.
Solutions of the Laplace equation are called potential functions or harmonic
functions. The Laplace equation is called also potential equation.
The general elliptic equation for a scalar function u(x), x R
n
, is
Lu :=
n
i,j=1
a
ij
(x)u
x
i
x
j
+
n
j=1
b
j
(x)u
x
j
+c(x)u = f(x),
where the matrix A = (a
ij
) is real, symmetric and positive denite. If A is
a constant matrix, then a transform to principal axis and stretching of axis
leads to
n
i,j=1
a
ij
u
x
i
x
j
= v,
where v(y) := u(Ty), T stands for the above composition of mappings.
7.1 Fundamental solution
Here we consider particular solutions of the Laplace equation in R
n
of the
type
u(x) = f([x y[),
175
176 CHAPTER 7. ELLIPTIC EQUATIONS OF SECOND ORDER
where y R
n
is xed and f is a function which we will determine such that
u denes a solution if the Laplace equation.
Set r = [x y[, then
u
x
i
= f
(r)
x
i
y
i
r
u
x
i
x
i
= f
(r)
(x
i
y
i
)
2
r
2
+f
(r)
_
1
r

(x
i
y
i
)
2
r
3
_
u = f
(r) +
n 1
r
f
(r).
Thus a solution of u = 0 is given by
f(r) =
_
c
1
ln r +c
2
: n = 2
c
1
r
2n
+c
2
: n 3
with constants c
1
, c
2
.
Denition. Set r = [x y[. The function
s(r) :=
_

1
2
ln r : n = 2
r
2n
(n2)
n
: n 3
is called singularity function associated to the Laplace equation. Here
is
n
the area of the n-dimensional unit sphere which is given by
n
=
2
n/2
/(n/2), where
(t) :=
_

0
e
t1
d, t > 0,
is the Gamma function.
Denition. A function
(x, y) = s(r) +(x, y)
is called fundamental solution associated to the Laplace equation if
C
2
() and
x
= 0 for each xed y .
Remark. The fundamental solution satises for each xed y the
relation
(x, y)
x
(x) dx = (y) for all C
2
0
(),
7.2. REPRESENTATION FORMULA 177
see an exercise. This formula follows from considerations similar to the next
section.
In the language of distribution, this relation can be written by denition
as
x
(x, y) = (x y),
where is the Dirac distribution, which is called -function.
7.2 Representation formula
In the following we assume that , the function which appears in the
denition of the fundamental solution and the potential function u consid-
ered are suciently regular such that the following calculations make sense,
see [6] for generalizations. This is the case if is bounded, is in C
1
,
C
2
() for each xed y and u C
2
().
x
y
n
x

Figure 7.1: Notations to Greens identity
Theorem 7.1. Let u be a potential function and a fundamental solution,
then for each xed y
u(y) =
_
_
(x, y)
u(x)
n
x
u(x)
(x, y)
n
x
_
dS
x
.
Proof. Let B
(y) be a ball. Set
(y) = B
(y). See Figure 7.2 for

notations. From Greens formula, for u, v C
2
(),
_
(y)
(vu uv) dx =
_
(y)
_
v
u
n
u
v
n
_
dS
y
n
(y)
Figure 7.2: Notations to Theorem 7.1

we obtain, if v is a fundamental solution and u a potential function,
_
(y)
_
v
u
n
u
v
n
_
dS = 0.
Thus we have to consider
_
(y)
v
u
n
dS =
_
v
u
n
dS +
_
B
(y)
v
u
n
dS
_
(y)
u
v
n
dS =
_
u
v
n
dS +
_
B
(y)
u
v
n
dS.
We estimate the integrals over B
(y):
(i)
_
B
(y)
v
u
n
dS
M
_
B
(y)
[v[ dS
M
_
_
B
(y)
s() dS +C
n
n1
_
,
where
M = M(y) = sup
B
0
(y)
[u/n[,
0
,
C = C(y) = sup
xB
0
(y)
[(x, y)[.
7.2. REPRESENTATION FORMULA 179
From the denition of s() we get the estimate as 0
_
B
(y)
v
u
n
dS =
_
O([ ln[) : n = 2
O() : n 3.
(7.1)
(ii) Consider the case n 3, then
_
B
(y)
u
v
n
dS =
1
n
_
B
(y)
u
1
n1
dS +
_
B
(y)
u
n
dS
=
1
n1
_
B
(y)
u dS +O(
n1
)
=
1
n1
u(x
0
)
_
B
(y)
dS +O(
n1
),
= u(x
0
) +O(
n1
).
for an x
0
B
(y).
Combining this estimate and (7.1), we obtain the representation formula of
the theorem. 2
Corollary. Set 0 and r = [x y[ in the representation formula of
Theorem 7.1, then
u(y) =
1
2
_
_
ln r
u
n
x
u
(ln r)
n
x
_
dS
x
, n = 2, (7.2)
u(y) =
1
(n 2)
n
_
_
1
r
n2
u
n
x
u
(r
2n
)
n
x
_
dS
x
, n 3. (7.3)
7.2.1 Conclusions from the representation formula
Similar to the theory of functions of one complex variable, we obtain here
results for harmonic functions from the representation formula, in particular
from (7.2), (7.3). We recall that a function u is called harmonic if u C
2
()
and u = 0 in .
Proposition 7.1. Assume u is harmonic in . Then u C
().
Proof. Let
0
be a domain such that y
0
. It follows from
representation formulas (7.2), (7.3), where :=
0
, that D
l
u(y) exist and
are continuous for all l since one can change dierentiation with integration
in right hand sides of the representation formulae. 2
Remark. In fact, a function which is harmonic in is even real analytic
in , see an exercise.
Proposition 7.2 (Mean value formula for harmonic functions). Assume u
is harmonic in . Then for each B
(x)
u(x) =
1
n1
_
B
(x)
u(y) dS
y
.
Proof. Consider the case n 3. The assertion follows from (7.3) where
:= B
(x) since r = and

_
B
(x)
1
r
n2
u
n
y
dS
y
=
1
n2
_
B
(x)
u
n
y
dS
y
=
1
n2
_
B
(x)
u dy
= 0.
2
We recall that a domain R
n
is called connected if is not the union of
two nonempty open subsets
1
,
2
such that
1
2
= . A domain in R
n
is connected if and only if its path connected.
Proposition 7.3 (Maximum principle). Assume u is harmonic in a con-
nected domain and achieves its supremum or inmum in . Then u const.
in .
Proof. Consider the case of the supremum. Let x
0
such that
u(x
0
) = sup
u(x) =: M.
Set
1
:= x : u(x) = M and
2
:= x : u(x) < M. The
set
1
is not empty since x
0

1
. The set
2
is open since u C
2
().
Consequently,
2
is empty if we can show that
1
is open. Let x
1
, then
there is a
0
> 0 such that B
0
(x) and u(x) = M for all x B
0
(x).
7.3. BOUNDARY VALUE PROBLEMS 181
If not, then there exists > 0 and x such that [ x x[ = , 0 < <
0
and
u( x) < M. From the mean value formula, see Proposition 7.2, it follows
M =
1
n1
_
B
(x)
u(x) dS <
M
n1
_
B
(x)
dS = M,
which is a contradiction. Thus, the set
2
is empty since
1
is open. 2
Corollary. Assume is connected and bounded, and u C
2
() C()
is harmonic in . Then u achieves its minimum and its maximum on the
boundary .
Remark. The previous corollary fails if is not bounded as simple coun-
terexamples show.
7.3 Boundary value problems
Assume R
n
is a connected domain.
7.3.1 Dirichlet problem
The Dirichlet problem (rst boundary value problem) is to nd a solution
u C
2
() C() of
u = 0 in (7.4)
u = on , (7.5)
where is given and continuous on .
Proposition 7.4. Assume is bounded, then a solution to the Dirichlet
problem is uniquely determined.
Proof. Maximum principle.
Remark. The previous result fails if we take away in the boundary condi-
tion (7.5) one point from the the boundary as the following example shows.
Let R
2
be the domain
= x B
1
(0) : x
2
> 0,
x
2
x
1
1
Figure 7.3: Counterexample
Assume u C
2
() C( 0) is a solution of
u = 0 in
u = 0 on 0.
This problem has solutions u 0 and u = Im(z +z
1
), where z = x
1
+ix
2
.
Another example see an exercise.
In contrast to this behaviour of the Laplace equation, one has uniqueness
if u = 0 is replaced by the minimal surface equation
x
1
_
u
x
1
_
1 +[u[
2
_
+

x
2
_
u
x
2
_
1 +[u[
2
_
= 0.
7.3.2 Neumann problem
The Neumann problem (second boundary value problem) is to nd a solution
u C
2
() C
1
() of
u = 0 in (7.6)
u
n
= on , (7.7)
where is given and continuous on .
Proposition 7.5. Assume is bounded, then a solution to the Dirichlet
problem is in the class u C
2
() uniquely determined up to a constant.
Proof. Exercise. Hint: Multiply the dierential equation w = 0 by w and
integrate the result over .
7.4. GREENS FUNCTION FOR 183
Another proof under the weaker assumption u C
1
() C
2
() follows
from the Hopf boundary point lemma, see Lecture Notes: Linear Elliptic
Equations of Second Order, for example.
7.3.3 Mixed boundary value problem
The Mixed boundary value problem (third boundary value problem) is to
nd a solution u C
2
() C
1
() of
u = 0 in (7.8)
u
n
+hu = on , (7.9)
where and h are given and continuous on .e and h are given and
continuous on .
Proposition 7.6. Assume is bounded and suciently regular, then a
solution to the mixed problem is uniquely determined in the class u C
2
()
provided h(x) 0 on and h(x) > 0 for at least one point x .
Proof. Exercise. Hint: Multiply the dierential equation w = 0 by w and
integrate the result over .
7.4 Greens function for
Theorem 7.1 says that each harmonic function satises
u(x) =
_
_
(y, x)
u(y)
n
y
u(y)
(y, x)
n
y
_
dS
y
, (7.10)
where (y, x) is a fundamental solution. In general, u does not satises the
boundary condition in the above boundary value problems. Since = s+,
see Section 7.2, where is an arbitrary harmonic function for each xed x,
we try to nd a such that u satises also the boundary condition.
Consider the Dirichlet problem, then we look for a such that
(y, x) = 0, y , x . (7.11)
Then
u(x) =
_
(y, x)
n
y
u(y) dS
y
, x .
Suppose that u achieves its boundary values of the Dirichlet problem,
then
u(x) =
_
(y, x)
n
y
(y) dS
y
, (7.12)
We claim that this function solves the Dirichlet problem (7.4), (7.5).
A function (y, x) which satises (7.11), and some additional assump-
tions, is called Greens function. More precisely, we dene a Green function
as follows.
Denition. A function G(y, x), y, x , x ,= y, is called Green function
associated to and to the Dirichlet problem (7.4), (7.5) if for xed x ,
that is we consider G(y, x) as a function of y, the following properties hold:
(i) G(y, x) C
2
( x) C( x),
y
G(y, x) = 0, x ,= y.
(ii) G(y, x) s([x y[) C
2
() C().
(iii) G(y, x) = 0 if y , x ,= y.
Remark. We will see in the next section that a Green function exists at
least for some domains of simple geometry. Concerning the existence of a
Green function for more general domains see [13].
It is an interesting fact that we get from (i)-(iii) of the above denition two
further important properties. We assume that is bounded, suciently
regular and connected.
Proposition 7.7. A Green function has the following properties. In the
case n = 2 we assume diam < 1.
(A) G(x, y) = G(y, x) (symmetry).
(B) 0 < G(x, y) < s([x y[), x, y , x ,= y.
Proof. (A) Let x
(1)
, x
(2)
. Set B
i
= B
(x
(i)
), i = 1, 2. We assume
B
i
and B
1
B
2
= . Since G(y, x
(1)
) and G(y, x
(2)
) are harmonic in
_
B
1
B
2
_
we obtain from Greens identity, see Figure 7.4 for notations,
n
n n
x
x
(1)
(2)
Figure 7.4: Proof of Proposition 7.7
0 =
_
(\(B
1
B
2
))
_
G(y, x
(1)
)

n
y
G(y, x
(2)
)
G(y, x
(2)
)

n
y
G(y, x
(1)
)
_
dS
y
=
_
_
G(y, x
(1)
)

n
y
G(y, x
(2)
) G(y, x
(2)
)

n
y
G(y, x
(1)
)
_
dS
y
+
_
B
1
_
G(y, x
(1)
)

n
y
G(y, x
(2)
) G(y, x
(2)
)

n
y
G(y, x
(1)
)
_
dS
y
+
_
B
2
_
G(y, x
(1)
)

n
y
G(y, x
(2)
) G(y, x
(2)
)

n
y
G(y, x
(1)
)
_
dS
y
.
The integral over is zero because of property (iii) of a Green function,
and
_
B
1
_
G(y, x
(1)
)

n
y
G(y, x
(2)
) G(y, x
(2)
)

n
y
G(y, x
(1)
)
_
dS
y
G(x
(1)
, x
(2)
),
_
B
2
_
G(y, x
(1)
)

n
y
G(y, x
(2)
) G(y, x
(2)
)

n
y
G(y, x
(1)
)
_
dS
y
G(x
(2)
, x
(1)
)
as 0. This follows by considerations as in the proof of Theorem 7.1.
(B) Since
G(y, x) = s([x y[) +(y, x)
and G(y, x) = 0 if y and x we have for y
(y, x) = s([x y[).
From the denition of s([x y[) it follows that (y, x) < 0 if y .
Thus, since
y
= 0 in , the maximum-minimum principle implies that
(y, x) < 0 for all y, x . Consequently
G(y, x) < s([x y[), x, y , x ,= y.
It remains to show that
G(y, x) > 0, x, y , x ,= y.
Fix x and let B
(x) be a ball such that B
(x) for all 0 < <

0
.
There is a suciently small
0
> 0 such that for each , 0 < <
0
,
G(y, x) > 0 for all y B
(x), x ,= y,
see property (iii) of a Green function. Since
y
G(y, x) = 0 in B
(x)
G(y, x) > 0 if y B
(x)
G(y, x) = 0 if y
it follows from the maximum-minimum principle that
G(y, x) > 0 on B
(x).
2
7.4.1 Greens function for a ball
If = B
R
(0) is a ball, then Greens function is explicitly known.
Let = B
R
(0) be a ball in R
n
with radius R and the center at the
origin. Let x, y B
R
(0) and let y
the reected point of y on the sphere

B
R
(0), that is, in particular [y[[y
[ = R
2
, see Figure 7.5 for notations. Set
G(x, y) = s(r) s
_
R
r
1
_
,
where s is the singularity function of Section 7.1, r = [x y[ and
2
=
n
i=1
y
2
i
, r
1
=
n
i=1
_
x
i

R
2
2
y
i
_
2
.
0
y
r
R
y
r
1
x
Figure 7.5: Reection on B
R
(0)
This function G(x, y) satises (i)-(iii) of the denition of a Green function.
We claim that
u(x) =
_
B
R
(0)
n
y
G(x, y) dS
y
is a solution of the Dirichlet problem (7.4), (7.5). This formula is also true
for a large class of domains R
n
, see [13].
Lemma.

n
y
G(x, y)
|y|=R
=
1
R
n
R
2
[x[
2
[y x[
n
.
Proof. Exercise.
Set
H(x, y) =
1
R
n
R
2
[x[
2
[y x[
n
, (7.13)
which is called Poissons kernel.
Theorem 7.2. Assume C(). Then
u(x) =
_
B
R
(0)
H(x, y)(y) dS
y
is the solution of the rst boundary value problem (7.4), (7.5) in the class
C
2
() C().
Proof. The proof follows from following properties of H(x, y):
(i) H(x, y) C
, [y[ = R, [x[ < R, x ,= y,

(ii)
x
H(x, y) = 0, [x[ < R, [y[ = R,
(iii)
_
B
R
(0)
H(x, y) dS
y
= 1, [x[ < R,
(iv) H(x, y) > 0, [y[ = R, [x[ < R,
(v) Fix B
R
(0) and > 0, then lim
x,|x|<R
H(x, y) = 0
uniformly in y B
R
(0), [y [ > .
(i), (iv) and (v) follow from the denition (7.13) of H and (ii) from (7.13)
or from
H =
G(x, y)
n
y
yB
R
(0)
,
G harmonic and G(x, y) = G(y, x).
Property (iii) is a consequence of formula
u(x) =
_
B
R
(0)
H(x, y)u(y) dS
y
,
for each harmonic function u, see calculations to the representation formula
above. We obtain (ii) if we set u 1.
It remains to show that u, given by Poissons formula, is in C(B
R
(0)) and
that u achieves the prescribed boundary values. Fix B
R
(0) and let
x B
R
(0). Then
u(x) () =
_
B
R
(0)
H(x, y) ((y) ()) dS
y
= I
1
+I
2
,
where
I
1
=
_
B
R
(0), |y|<
H(x, y) ((y) ()) dS
y
I
2
=
_
B
R
(0), |y|
H(x, y) ((y) ()) dS
y
.
For given (small) > 0 there is a = () > 0 such that
[(y) ()[ <
for all y B
R
(0) with [y [ < . It follows [I
1
[ because of (iii) and
(iv).
Set M = max
B
R
(0)
[[. From (v) we conclude that there is a
> 0 such
that
H(x, y) <

2M
n
R
n1
if x and y satisfy [x[ <
, [y [ > , see Figure 7.6 for notations. Thus

R
x
Figure 7.6: Proof of Theorem 7.2

[I
2
[ < and the inequality
[u(x) ()[ < 2
for x B
R
(0) such that [x [ <
is shown. 2
Remark. Dene [0, ] through cos = x y/([x[[y[), then we write
Poissons formula of Theorem 7.2 as
u(x) =
R
2
[x[
2
n
R
_
B
R
(0)
(y)
1
([x[
2
+R
2
2[x[Rcos )
n/2
dS
y
.
In the case n = 2 we can expand this integral in a power series with respect
to := [x[/R if [x[ < R, since
R
2
[x[
2
[x[ +R
2
2[x[Rcos
=
1
2
2
2 cos + 1
= 1 + 2
n=1
n
cos(n),
see [16], pp. 18 for an easy proof of this formula, or [4], Vol. II, p. 246.
7.4.2 Greens function and conformal mapping
For two-dimensional domains there is a beautiful connection between confor-
mal mapping and Greens function. Let w = f(z) be a conformal mapping
from a suciently regular connected domain in R
2
onto the interior of the
unit circle, see Figure 7.7. Then the Green function of is, see for exam-
w=f(z)
z
w
x
y
1
.
.
E
Figure 7.7: Conformal mapping
ple [16] or other text books about the theory of functions of one complex
variable,
G(z, z
0
) =
1
2
ln
1 f(z)f(z
0
)
f(z) f(z
0
)
,
where z = x
1
+ix
2
, z
0
= y
1
+iy
2
.
7.5 Inhomogeneous equation
Here we consider solutions u C
2
() C() of
u = f(x) in (7.14)
u = 0 on , (7.15)
where f is given.
We need the following lemma concerning volume potentials. We assume
that is bounded and suciently regular such that all the following integrals
exist. See [6] for generalizations concerning these assumptions.
Let for x R
n
, n 3,
V (x) =
_
f(y)
1
[x y[
n2
dy
and set in the two-dimensional case
V (x) =
_
f(y) ln
_
1
[x y[
_
dy.
We recall that
n
= [B
1
(0)[.
Lemma.
(i) Assume f C(). Then V C
1
(R
n
) and
V
x
i
(x) =
_
f(y)

x
i
_
1
[x y[
n2
_
dy, if n 3,
V
x
i
(x) =
_
f(y)

x
i
_
ln
_
1
[x y[
__
dy if n = 2.
(ii) If f C
1
(), then V C
2
() and
V = (n 2)
n
f(x), x , n 3
V = 2f(x), x , n = 2.
Proof. To simplify the presentation, we consider the case n = 3.
(i) The rst assertion follows since we can change dierentiation with inte-
gration since the dierentiate integrand is weakly singular, see an exercise.
(ii) We will dierentiate at x . Let B
be a xed ball such that x B
,
suciently small such that B
. Then, according to (i) and since we

have the identity
x
i
_
1
[x y[
_
=

y
i
_
1
[x y[
_
which implies that
f(y)

x
i
_
1
[x y[
_
=

y
i
_
f(y)
1
[x y[
_
+f
y
i
(y)
1
[x y[
,
we obtain
V
x
i
(x) =
_
f(y)

x
i
_
1
[x y[
_
dy
=
_
\B
f(y)

x
i
_
1
[x y[
_
dy +
_
B
f(y)

x
i
_
1
[x y[
_
dy
=
_
\B
f(y)

x
i
_
1
[x y[
_
dy
+
_
B

y
i
_
f(y)
1
[x y[
_
+f
y
i
(y)
1
[x y[
_
dy
=
_
\B
f(y)

x
i
_
1
[x y[
_
dy
+
_
B
f
y
i
(y)
1
[x y[
dy
_
B
f(y)
1
[x y[
n
i
dS
y
,
where n is the exterior unit normal at B
. It follows that the rst and

second integral is in C
1
(). The second integral is also in C
1
() according
to (i) and since f C
1
() by assumption.
Because of
x
([x y[
1
) = 0, x ,= y, it follows
V =
_
B
i=1
f
y
i
(y)

x
i
_
1
[x y[
_
dy
_
B
f(y)
n
i=1
x
i
_
1
[x y[
_
n
i
dS
y
.
Now we choose for B
a ball with the center at x, then

V = I
1
+I
2
,
where
I
1
=
_
B
(x)
n
i=1
f
y
i
(y)
y
i
x
i
[x y[
3
dy
I
2
=
_
B
(x)
f(y)
1
2
dS
y
.
We recall that n (y x) = if y B
(x). It is I
1
= O() as 0 and
for I
2
we obtain from the mean value theorem of the integral calculus that
for a y B
(x)
I
2
=
1
2
f(y)
_
B
(x)
dS
y
=
n
f(y),
which implies that lim
0
I
2
=
n
f(x). 2
In the following we assume that Greens function exists for the domain ,
which is the case if is a ball.
Theorem 7.3. Assume f C
1
() C(). Then
u(x) =
_
G(x, y)f(y) dy
is the solution of the inhomogeneous problem (7.14), (7.15).
Proof. For simplicity of the presentation let n = 3. We will show that
u(x) :=
_
G(x, y)f(y) dy
is a solution of (7.4), (7.5). Since
G(x, y) =
1
4[x y[
+(x, y),
where is a potential function with respect to x or y, we obtain from the
above lemma that
u =
1
4
f(y)
1
[x y[
dy +
_
x
(x, y)f(y) dy
= f(x),
where x . It remains to show that u achieves its boundary values. That
is, for xed x
0
we will prove that
lim
xx
0
, x
u(x) = 0.
Set
u(x) = I
1
+I
2
,
where
I
1
(x) =
_
\B
(x
0
)
G(x, y)f(y) dy,
I
2
(x) =
_
B
(x
0
)
G(x, y)f(y) dy.
Let M = max
[f(x)[. Since
G(x, y) =
1
4
1
[x y[
+(x, y),
we obtain, if x B
(x
0
) ,
[I
2
[
M
4
_
B
(x
0
)
dy
[x y[
+O(
2
)
M
4
_
B
2(x)
dy
[x y[
+O(
2
)
= O(
2
)
as 0. Consequently for given there is a
0
=
0
() > 0 such that
[I
2
[ <

2
for all 0 <
0
.
For each xed , 0 <
0
, we have
lim
xx
0
, x
I
1
(x) = 0
since G(x
0
, y) = 0 if y B
(x
0
) and G(x, y) is uniformly continuous in
x B
/2
(x
0
) and y B
(x
0
), see Figure 7.8. 2
Remark. For the proof of (ii) in the above lemma it is sucient to assume
that f is Holder continuous. More precisely, let f C
(), 0 < < 1, then

V C
2,
(), see for example [9].
7.6. EXERCISES 195
y
x
x
0
Figure 7.8: Proof of Theorem 7.3
7.6 Exercises
1. Let (x, y) be a fundamental solution to , y . Show that
(x, y) (x) dx = (y) for all C

2
0
() .
Hint: See the proof of the representation formula.
2. Show that [x[
1
sin(k[x[) is a solution of the Helmholtz equation
u +k
2
u = 0 in R
n
0.
3. Assume u C
2
(), bounded and suciently regular, is a solution
of
u = u
3
in
u = 0 on .
Show that u = 0 in .
4. Let
= x R
2
: x
1
> 0, 0 < x
2
< x
1
tan , 0 < . Show
that
u(x) = r
k
sin
_
k
_
is a harmonic function in
satisfying u = 0 on
, provided k is
an integer. Here (r, ) are polar coordinates with the center at (0, 0).
5. Let u C
2
() be a solution of u = 0 on the quadrangle =
(0, 1) (0, 1) satisfying the boundary conditions u(0, y) = u(1, y) = 0
for all y [0, 1] and u
y
(x, 0) = u
y
(x, 1) = 0 for all x [0, 1]. Prove
that u 0 in .
6. Let u C
2
(R
n
) be a solution of u = 0 in R
n
satisfying u L
2
(R
n
),
i. e.,
_
R
n
u
2
(x) dx < .
Show that u 0 in R
n
.
Hint: Prove
_
B
R
(0)
[u[
2
dx
const.
R
2
_
B
2R
(0)
[u[
2
dx,
where c is a constant independent of R.
To show this inequality, multiply the dierential equation by :=
2
u,
where C
1
is a cut-o function with properties: 1 in B
R
(0),
0 in the exterior of B
2R
(0), 0 1, [[ C/R. Integrate
the product, apply integration by parts and use the formula 2ab
a
2
+
1
b
2
, > 0.
7. Show that a bounded harmonic function dened on R
n
must be a
constant (a theorem of Liouville).
8. Assume u C
2
(B
1
(0)) C(B
1
(0) (1, 0)) is a solution of
u = 0 in B
1
(0)
u = 0 on B
1
(0) (1, 0).
Show that there are at least two solutions.
Hint: Consider
u(x, y) =
1 (x
2
+y
2
)
(1 x)
2
+y
2
.
9. Assume R
n
is bounded and u, v C
2
()C() satisfy u = v
and max
[u v[ for given > 0. Show that max
[u v[ .
10. Set = R
n
B
1
(0) and let u C
2
() be a harmonic function in
satisfying lim
|x|
u(x) = 0. Prove that
max
[u[ = max
[u[ .
Hint: Apply the maximum principle to B
R
(0), R large.
7.6. EXERCISES 197
11. Let
= x R
2
: x
1
> 0, 0 < x
2
< x
1
tan , 0 < ,
,R
=
B
R
(0), and assume f is given and bounded on
,R
.
Show that for each solution u C
1
(
,R
) C
2
(
,R
) of u = f in
,R
satisfying u = 0 on
,R
B
R
(0), holds:
For given > 0 there is a constant C() such that
[u(x)[ C() [x[
in
,R
.
Hint: (a) Comparison principle (a consequence from the maximum
principle): Assume is bounded, u, v C
2
() C() satisfying
u v in and u v on . Then u v in .
(b) An appropriate comparison function is
v = Ar
sin(B( +)) ,
A, B, appropriate constants, B, positive.
12. Let be the quadrangle (1, 1) (1, 1) and u C
2
() C() a
solution of the boundary value problem u = 1 in , u = 0 on .
Find a lower and an upper bound for u(0, 0).
Hint: Consider the comparison function v = A(x
2
+y
2
), A = const.
13. Let u C
2
(B
a
(0)) C(B
a
(0)) satisfying u 0, u = 0 in B
a
(0).
Prove (Harnacks inequality):
a
n2
(a [[)
(a +[[)
n1
u(0) u()
a
n2
(a +[[)
(a [[)
n1
u(0) .
Hint: Use the formula (see Theorem 7.2)
u(y) =
a
2
[y[
2
a
n
_
|x|=a
u(x)
[x y[
n
dS
x
for y = and y = 0.
14. Let () be a 2-periodic C
4
-function with the Fourier series
() =
n=0
(a
n
cos(n) +b
n
sin(n)) .
Show that
u =
n=0
(a
n
cos(n) +b
n
sin(n)) r
n
solves the Dirichlet problem in B
1
(0).
15. Assume u C
2
() satises u = 0 in . Let B
a
() be a ball such
that its closure is in . Show that
[D
u()[ M
_
[[
n
a
_
||
,
where M = sup
xB
a
()
[u(x)[ and
n
= 2n
n1
/((n 1)
n
).
Hint: Use the formula of Theorem 7.2, successively to the k th deriva-
tives in balls with radius a([[ k)/m, k = o, 1, . . . , m1.
16. Use the result of the previous exercise to show that u C
2
() satis-
fying u = 0 in is real analytic in .
Hint: Use Stirlings formula
n! = n
n
e
n
_
2n +O
_
1
n
__
as n , to show that u is in the class C
K,r
(), where K = cM and
r = a/(e
n
). The constant c is the constant in the estimate n
n
ce
n
n!
which follows from Stirlings formula. See Section 3.5 for the denition
of a real analytic function.
17. Assume is connected and u C
2
() is a solution of u = 0 in .
Prove that u 0 in if D
u() = 0 for all , for a point . In

particular, u 0 in if u 0 in an open subset of .
18. Let = (x
1
, x
2
, x
3
) R
3
: x
3
> 0, which is a half-space of R
3
.
Show that
G(x, y) =
1
4[x y[

1
4[x y[
,
where y = (y
1
, y
2
, y
3
), is the Green function to .
19. Let = (x
1
, x
2
, x
3
) R
3
: x
2
1
+x
2
2
+x
2
3
< R
2
, x
3
> 0, which is half
of a ball in R
3
. Show that
G(x, y) =
1
4[x y[

R
4[y[[x y
1
4[x y[
+
R
4[y[[x y
[
,
where y = (y
1
, y
2
, y
3
), y
= R
2
y/([y[
2
) and y
= R
2
y/([y[
2
), is the
Green function to .
7.6. EXERCISES 199
20. Let = (x
1
, x
2
, x
3
) R
3
: x
2
> 0, x
3
> 0, which is a wedge in R
3
.
Show that
G(x, y) =
1
4[x y[

1
4[x y[
1
4[x y
[
+
1
4[x y
[
,
where y = (y
1
, y
2
, y
3
), y
= (y
1
, y
2
, y
3
) and y
= (y
1
, y
2
, y
3
), is
the Green function to .
21. Find Greens function for the exterior of a disk, i. e., of the domain
= x R
2
: [x[ > R.
22. Find Greens function for the angle domain = z C : 0 < arg z <
, 0 < < .
23. Find Greens function for the slit domain = z C : 0 < arg z <
2.
24. Let for a suciently regular domain R
n
, a ball or a quadrangle
for example,
F(x) =
_
K(x, y) dy,
where K(x, y) is continuous in where x ,= y, and which satises
[K(x, y)[
c
[x y[
with a constants c and , < n.

Show that F(x) is continuous on .
25. Prove (i) of the lemma of Section 7.5.
Hint: Consider the case n 3. Fix a function C
1
(R) satisfying
0 1, 0
2, (t) = 0 for t 1, (t) = 1 for t 2 and

consider for > 0 the regularized integral
V
(x) :=
_
f(y)
dy
[x y[
n2
,
where
= ([x y[/). Show that V
converges uniformly to V
on compact subsets of R
n
as 0, and that V
(x)/x
i
converges
uniformly on compact subsets of R
n
to
_
f(y)

x
i
_
1
[x y[
n2
_
dy
as 0.
26. Consider the inhomogeneous Dirichlet problem u = f in , u =
on . Transform this problem into a Dirichlet problem for the
Laplace equation.
Hint: Set u = w +v, where w(x) :=
_
s([x y[)f(y) dy.

Bibliography
[1] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
with Formulas, Graphs, and Mathematical tables. Vol. 55, National Bu-
reau of Standards Applied Mathematics Series, U.S. Government Print-
ing Oce, Washington, DC, 1964. Reprinted by Dover, New York, 1972.
[2] S. Bernstein, Sur un theor`eme de geometrie et son application aux
derivees partielles du type elliptique. Comm. Soc. Math. de Kharkov
(2)15,(19151917), 3845. German translation: Math. Z. 26 (1927),
551558.
[3] E. Bombieri, E. De Giorgi and E. Giusti, Minimal cones and the Bern-
stein problem. Inv. Math. 7 (1969), 243268.
[4] R. Courant und D. Hilbert, Methoden der Mathematischen Physik.
Band 1 und Band 2. Springer-Verlag, Berlin, 1968. English translation:
Methods of Mathematical Physics. Vol. 1 and Vol. 2, Wiley-Interscience,
1962.
[5] L. C. Evans, Partial Dierential Equations. Graduate Studies in Math-
ematics , Vol. 19, AMS, Providence, 1991.
[6] L. C. Evans and R. F. Gariepy, Measure Theory and Fine Properties of
Functions, Studies in Advanced Mathematics, CRC Press, Boca Raton,
1992.
[7] R. Finn, Equilibrium Capillary Surfaces. Grundlehren, Vol. 284,
Springer-Verlag, New York, 1986.
[8] P. R. Garabedian, Partial Dierential Equations. Chelsia Publishing
Company, New York, 1986.
[9] D. Gilbarg and N. S. Trudinger, Elliptic Partial Dierential Equations
of Second Order. Grundlehren, Vol. 224, Springer-Verlag, Berlin, 1983.
201
202 BIBLIOGRAPHY
[10] F. John, Partial Dierential Equations. Springer-Verlag, New York,
1982.
[11] K. Konigsberger, Analysis 2. Springer-Verlag, Berlin, 1993.
[12] L. D. Landau and E. M. Lifschitz, Lehrbuch der Theoretischen Physik.
Vol. 1., Akademie-Verlag, Berlin, 1964. German translation from Rus-
sian. English translation: Course of Theoretical Physics. Vol. 1, Perga-
mon Press, Oxford, 1976.
[13] R. Leis, Vorlesungen uber partielle Dierentialgleichungen zweiter Ord-
nung. B. I.-Hochschultaschenb ucher 165/165a, Mannheim, 1967.
[14] J.-L. Lions and E. Magenes, Problemes aux limites non homogenes et
applications. Dunod, Paris, 1968.
[15] E. Miersemann, Kapillar achen. Ber. Verh. Sachs. Akad. Wiss. Leipzig,
Math.-Natur. Kl. 130 (2008), Heft 4, S. Hirzel, Leipzig, 2008.
[16] Z. Nehari, Conformal Mapping. Reprinted by Dover, New York, 1975.
[17] I. G. Petrowski, Vorlesungen uber Partielle Dierentialgleichungen.
Teubner, Leipzig, 1955. Translation from Russian. Englisch translation:
Lectures on Partial Dierential Equations. Wiley-Interscience, 1954.
[18] H. Sagan, Introduction to the Calculus of Variations. Dover, New York,
1992.
[19] J. Simons, Minimal varieties in riemannian manifolds. Ann. of Math(2)
88 (1968), 62105.
[20] W. I. Smirnow, Lehrgang der H oheren Mathematik., Teil II. VEB Verlag
der Wiss., Berlin, 1975. Translation from Russian. English translation:
Course of Higher Mathematics, Vol. 2., Elsevier, 1964.
[21] W. I. Smirnow, Lehrgang der H oheren Mathematik., Teil IV. VEB Ver-
lag der Wiss., Berlin, 1975. Translation from Russian. English transla-
tion: Course of Higher Mathematics, Vol. 4., Elsevier, 1964.
[22] A. Sommerfeld, Partielle Dierentialgleichungen. Geest & Portig,
Leipzig, 1954.
[23] W. A. Strauss, Partial Dierential equations. An Introduction. Second
edition, Wiley-Interscience, 2008. German translation: Partielle Dier-
entialgleichungen. Vieweg, 1995.
BIBLIOGRAPHY 203
[24] M. E. Taylor, Pseudodierential operators. Princeton, New Jersey, 1981.
[25] G. N. Watson, A treatise on the Theory of Bessel Functions. Cambridge,
1952.
[26] P. Wilmott, S. Howison and J. Dewynne, The Mathematics of Financial
Derivatives, A Student Introduction. Cambridge University Press, 1996.
[27] K. Yosida, Functional Analysis. Grundlehren, Vol. 123, Springer-Verlag,
Berlin, 1965.
Index
dAlembert formula 108
asymptotic expansion 84
basic lemma 16
Beltrami equations
Black-Scholes equation 164
Black-Scholes formulae 165, 169
boundary condition 14, 15
capillary equation 21
Cauchy-Kowalevskaya theorem 63,
84
Cauchy-Riemann equations 13
characteristic equation 28, 33, 41,
47, 74
characteristic curve 28
characteristic strip 47
classication
linear equations second order
63
quasilinear equations second or-
der 73
systems rst order 74
cylinder surface 29
diusion 163
Dirac distribution 177
Dirichlet integral 17
Dirichlet problem 181
domain of dependence 108, 115
domain of inuence 109
elliptic 73, 75
nonuniformly elliptic 73
second order 175
system 75, 82
uniformly elliptic 73
Euler-Poisson-Darboux equation 111
Euler equation 15, 17
rst order equations 25
two variables 40
R
n
51
Fourier transform 141
inverse Fourier transform 142
Fouriers method 126, 162
functionally dependent 13
fundamental solution 175, 176
Gamma function 176
gas dynamics 79
Greens function 183
ball 186
conformal mapping 190
Hamilton function 54
Hamilton-Jacobi theory 53
harmonic function 179
heat equation 14
inhomogeneous 155
heat kernel 152, 153
helicoid 30
hyperbolic
equation 107
inhomogeneous equation 117
one dimensional 107
higher dimension 109
system 74
initial conditions 15
initial-boundary value problem
uniqueness 134
204
INDEX 205
string 125
membrane 128, 129
initial value problem
Cauchy 33, 48
integral of a system 28
Jacobi theorem 55
Kepler 56
Laplace equation 13, 20
linear elasticity 83
linear equation 11, 25
second order 63
maximum principle
heat equation 156
parabolic 161
harmonic function 180
Maxwell equations 76
mean value formula 180
minimal surface equation 18
Monge cone 42, 43
multi-index 90
multiplier 12
Navier Stokes 83
Neumann problem 20, 182
Newton potential 13
normal form 69
noncharacteristic curve 33
option
call 165
put 169
parabolic
equation 151
system 75
Picard-Lindelof theorem 9
Poissons formula 152
Poissons kernel 187
pseudodierential operators 146, 147
quasiconform mapping 76
quasilinear equation 11, 31
real analytic function 90
resonance 134
Riemanns method 120
Riemann problem 61
Schrodinger equation 137
separation of variables 126
singularity function 176
speed
plane 78
relative 81
surface 81, 83
sound 82
spherical mean 110
strip condition 46
telegraph equation 78
wave equation 14, 107, 131
wave front 51
volume potential 191

Pde Book

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pde Book

Uploaded by

Copyright:

Available Formats

Partial Dierential Equations

(x) = f(x, y(x))

(x) = f(x, y(x), y

(x) = f(x, y(x), y

(x) = f(x, y(x), y

= 0 and vice versa, thus v = w() are solutions for

(0) and the following basic lemma

Figure 1.6: Comparison surface

Figure 2.4: Cauchy initial value problem

) in a neighbourhood of the curve (x

, t), y(t) := y(s

, t), z(t) := z(s

(t) from the characteristic equations

Figure 2.13: Motion in a central eld

contains the straight line (0, 0, 1) +s(1, 1, 0), s R.

(x) = f(x, y(x)), y(x

from the (known) coecients

= lower order terms, (3.11)

(s) ,= 0, n is a unit vector and V = c/

(t) is the speed of plane waves

() > 0 if > 0. Then the above system of four equations is

() is called speed of sound .

with respect to their arguments. From (3.26)

are polynomials in the indicated arguments with

have integers as coecients. The initial condition (3.29)

(x) = f(x, y(x))

u(y +(x y))) x

u(y +t(x y)) dt

(D) and the formal dierentiate series, that is

R and if there is a neighbourhood

() in the case that f is real analytic in the domain .

(). Then for each y there exists a neighbourhood

() and is connected. Then f is

f(z) are known.

f(x) = 0 for all ,

f(x) ,= 0 for at least one .

f are continuous in . The set

is real analytic if these estimates are satised.

() and for every

converges (absolutely) in a neighbourhood of y and that this series is equal

f(y +t(x y))(x y)

are polynomials with nonnegative integers as coecients, P

(u) : = g(v +u) C

(x) : = f(y +x) f(y) C

(x) = f(x, y(x))

(x) = (x). (4.5)

(ct) +F(ct) +tG(ct)

where < 0, see Figure 4.5. We have S = S

is given by, see Theorem

(t) (strip condition),

(t) = 0, see Chapter 2.

(vLu uMv) dxdy =

4cs leads to Bessels dierential equation

(x) = v(x) in (0, l) (4.50)

r and v(r) = v(z/

) =: y(z), then, see (4.64),

(x) = v(x) in (0, l),

f() d = f(x) (5.3)

G(y)f(x +y) dy.

are given suciently

[0, ) and a solution of the initial value problem (6.1), (6.2).

(x, t) : = u(x, t) (4(T + t))

(y), see Figure 6.3. On the

(x, 0) u(x, 0) sup