Partial Differential Equations - LibreTexts

PARTIAL DIFFERENTIAL
EQUATIONS
Erich Miersemann
Universität Leipzig
Universität Leipzig
Partial Differential Equations
Erich Miersemann
This text is disseminated via the Open Education Resource (OER) LibreTexts Project (https://LibreTexts.org) and like the hundreds
of other texts available within this powerful platform, it is freely available for reading, printing and "consuming." Most, but not all,
pages in the library have licenses that may allow individuals to make changes, save, and print this book. Carefully
consult the applicable license(s) before pursuing such effects.
Instructors can adopt existing LibreTexts texts or Remix them to quickly build course-specific resources to meet the needs of their
students. Unlike traditional textbooks, LibreTexts’ web based origins allow powerful integration of advanced features and new
technologies to support learning.
The LibreTexts mission is to unite students, faculty and scholars in a cooperative effort to develop an easy-to-use online platform
for the construction, customization, and dissemination of OER content to reduce the burdens of unreasonable textbook costs to our
students and society. The LibreTexts project is a multi-institutional collaborative venture to develop the next generation of open-
access texts to improve postsecondary education at all levels of higher learning by developing an Open Access Resource
environment. The project currently consists of 14 independently operating and interconnected libraries that are constantly being
optimized by students, faculty, and outside experts to supplant conventional paper-based books. These free textbook alternatives are
organized within a central environment that is both vertically (from advance to basic level) and horizontally (across different fields)
integrated.
The LibreTexts libraries are Powered by MindTouch® and are supported by the Department of Education Open Textbook Pilot
Project, the UC Davis Office of the Provost, the UC Davis Library, the California State University Affordable Learning Solutions
Program, and Merlot. This material is based upon work supported by the National Science Foundation under Grant No. 1246120,
1525057, and 1413739. Unless otherwise noted, LibreTexts content is licensed by CC BY-NC-SA 3.0.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not
necessarily reflect the views of the National Science Foundation nor the US Department of Education.
Have questions or comments? For information about adoptions or adaptions contact info@LibreTexts.org. More information on our
activities can be found via Facebook (https://facebook.com/Libretexts), Twitter (https://twitter.com/libretexts), or our blog
(http://Blog.Libretexts.org).
This text was compiled on 03/30/2022

TABLE OF CONTENTS
Partial differential equations are differential equations that contains unknown multivariable functions and their partial derivatives
1: INTRODUCTION
1.0: INTRODUCTION
1.1: EXAMPLES
1.2: ORDINARY DIFFERENTIAL EQUATIONS
1.3: PARTIAL DIFFERENTIAL EQUATIONS
1.E: INTRODUCTION (EXERCISES)
2: EQUATIONS OF FIRST ORDER

The main tool for studying related problems is the theory of ordinary differential equations, which is quite different for systems of partial
differential of first order.
2.0: PRELUDE TO FIRST ORDER EQUATIONS

2.1: LINEAR EQUATIONS
2.2: QUASILINEAR EQUATIONS
2.2.1: A LINEARIZATION METHOD
2.2.2: INITIAL VALUE PROBLEM OF CAUCHY
2.3: NONLINEAR EQUATIONS IN TWO VARIABLES
2.4: NONLINEAR EQUATIONS IN R
n
2.5: HAMILTON-JACOBI THEORY

2.E: EQUATIONS OF FIRST ORDER (EXERCISES)
3: CLASSIFICATION
3.1: LINEAR EQUATIONS OF SECOND ORDER
3.1.1: NORMAL FORM IN TWO VARIABLES
3.2: QUASILINEAR EQUATIONS OF SECOND ORDER
3.2.1: QUASILINEAR ELLIPTIC EQUATIONS
3.3: SYSTEMS OF FIRST ORDER
3.3.1: EXAMPLES
3.4: SYSTEMS OF SECOND ORDER
3.4.1: EXAMPLES
3.5: THEOREM OF CAUCHY-KOVALEVSKAYA
3.5.1 APPENDIX: REAL ANALYTIC FUNCTIONS
3.E: CLASSIFICATION (EXERCISES)
BACK MATTER
INDEX
4: HYPERBOLIC EQUATIONS
Here we consider hyperbolic equations of second order, mainly wave equations.
4.1: ONE-DIMENSIONAL WAVE EQUATION

4.2: HIGHER DIMENSIONS
4.2.1: CASE N=3
4.2.2: CASE N=2
4.3: INHOMOGENEOUS EQUATIONS
4.4: A METHOD OF RIEMANN
4.5: INITIAL-BOUNDARY VALUE PROBLEMS
4.5.1: OSCILLATION OF A STRING
4.5.2: OSCILLATION OF A MEMBRANE
4.5.3: INHOMOGENEOUS WAVE EQUATIONS
4.E: 4: HYPERBOLIC EQUATIONS (EXERCISES)
BACK MATTER
1 3/30/2022
INDEX
5: FOURIER TRANSFORM
5.1: DEFINITION AND PROPERTIES
5.1.1: PSEUDODIFFERENTIAL OPERATORS
5.E: FOURIER TRANSFORM (EXERCISES)
6: PARABOLIC EQUATIONS
Here we consider linear parabolic equations of second order.
6.1: POISSON'S FORMULA

6.2: INHOMOGENEOUS HEAT EQUATION
6.3: MAXIMUM PRINCIPLE
6.4.1: FOURIER'S METHOD
6.4.2: UNIQUENESS
6.5: BLACK-SCHOLES EQUATION
6.E: PARABOLIC EQUATIONS (EXERCISES)
7: ELLIPTIC EQUATIONS OF SECOND ORDER

7.1: FUNDAMENTAL SOLUTION
7.2: REPRESENTATION FORMULA
7.2.1: CONCLUSIONS FROM THE REPRESENTATION FORMULA
7.3.1: BOUNDARY VALUE PROBLEMS: DIRICHLET PROBLEM
7.3.2: BOUNDARY VALUE PROBLEMS: NEUMANN PROBLEM
7.3.3: BOUNDARY VALUE PROBLEMS: MIXED BOUNDARY VALUE PROBLEM
7.4: GREEN'S FUNCTION FOR Δ
7.4.1: GREEN'S FUNCTION FOR A BALL
7.4.2: GREEN'S FUNCTION AND CONFORMAL MAPPING
7.5: INHOMOGENEOUS EQUATION
7.E: ELLIPTIC EQUATIONS OF SECOND ORDER (EXERCISES)
BIBLIOGRAPHY
BACK MATTER
INDEX
GLOSSARY
2 3/30/2022
CHAPTER OVERVIEW
1: INTRODUCTION
Thumbnail: Visualization of heat transfer in a pump casing, created by solving the heat equation. Heat is being generated internally in the
casing and being cooled at the boundary, providing asteady state temperature distribution. (CC-BY-SA-3.0; Wikipedia A1).
1.0: INTRODUCTION
1.1: EXAMPLES
1.2: ORDINARY DIFFERENTIAL EQUATIONS
1.3: PARTIAL DIFFERENTIAL EQUATIONS
1.E: INTRODUCTION (EXERCISES)
1 3/30/2022
1.0: Introduction
Preface
These lecture notes are intented as a straightforward introduction to partial differential 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.
Introduction
Ordinary and partial differential equations occur in many applications. An ordinary differential equation is a special case of a
partial differential equation but the behavior of solutions is quite different in general 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),\ldots,y^{(n)})=0\]
is an ordinary differential equation of n-th order for the unknown function y(x), where F is given.
An important problem for ordinary differential equations is the initial value problem
′
y (x) = f (x, y(x))
y(x0 ) = y0 ,
where f is a given real function of two variables x,y and x ,, y are given real numbers.
0 0
Picard-Lindelöf Theorem. Suppose

(i) f (x, y) is continuous in a rectangle
2
Q = {(x, y) ∈ R : |x − x0 | < a, |y − y0 | < b}. (1.0.1)
(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, y2 ) − f (x, y1 )| ≤ L| y2 − y1 | (1.0.2)
for all (x, y1 ), (x, y2 ) .
Figure 1.0.1: Initial value problem

Then there exists a unique solution y ∈ C 1
(x0 − α, x0 + α) of the above initial value problem, where α = min(b/K, a) .
The linear ordinary differential equation
$$y^{(n)}+a_{n-1}(x)y^{(n-1)}+\ldots a_1(x)y'+a_0(x)y=0,\]
where a are continuous functions, has exactly n linearly independent solutions. In contrast to this property the partial differential
j
= 0 in R has infinitely many linearly independent solutions in the linear space C (R ) .

2 2 2
uxx+u yy
The ordinary differential equation of second order
Erich Miersemann 1.0.1 3/30/2022 https://math.libretexts.org/@go/page/2193

$$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(x0 ) = y0 , y (x0 ) = y1 ,
where y and y are given, or to consider the boundary value problem

0 1
′′ ′
y (x) = f (x, y(x), y (x))
y(x0 ) = y0 , y(x1 ) = y1 .
Figure 1.0.2: Boundary value problem

Initial and boundary value problems play an important role also in the theory of partial differential equations. A partial differential
equation for 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 derivatives 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 first order.
An equation is said to be quasilinear if it is linear in the highest derivatives. 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.
Contributors and Attributions

Prof. Dr. Erich Miersemann (Universität Leipzig)
Integrated by Justin Marshall.
Contributions and Attributions

This page is licensed under a not declared license and was authored, remixed, and/or curated by Erich Miersemann. A detailed
versioning history of the edits to source content is available upon request.

1.1: Examples
Example 1.1.1:
uy = 0 , where u = u(x, y). All functions u = w(x) are solutions.
Example 1.1.2:
ux = uy , where u = u(x, y) . A change of coordinates transforms this equation into an equation of the first example. Set
ξ = x + y, η = x − y , then
ξ +η ξ −η
u(x, y) = u ( , ) =: v(ξ, η). (1.1.1)
2 2
Assume u ∈ C , then
1
1
vη = (ux − uy ). (1.1.2)
2
If u = u , then v = 0 and vice versa, thus v = w(ξ) are solutions for arbitrary C -functions w(ξ). Consequently, we have a
x y η
1
large class of solutions of the original partial differential equation: u = w(x + y) with an arbitrary C -function w.
1
Example 1.1.3:
A necessary and sufficient condition such that for given C -functions M , N the integral
1
P1
∫ M (x, y)dx + N (x, y)dy (1.1.3)

P0
is independent of the curve which connects the points P0 with P1 in a simply connected domain Ω ⊂R
2
is that the partial
differential equation (condition of integrability)
My = Nx (1.1.4)
in Ω.
Figure 1.1.1: Independence of the path

This is one equation for two functions. A large class of solutions is given by M = Φ , N = Φ , where Φ(x, y) is an arbitrary
x y
C -function. It follows from Gauss theorem that these are all C -solutions of the above differential equation.
2 1
Example 1.1.4: Method of an integrating multiplier for an ordinary

differential

equation
Consider the ordinary differential equation
M (x, y)dx + N (x, y)dy = 0 (1.1.5)
for given C -functions M , N . Then we seek a C -function μ(x, y) such that μM dx + μN dy is a total differential, i. e., that
1 1
(μM ) = (μN )
y xis satisfied. This is a linear partial differential equation of first order for μ :
$$
M\mu_y-N\mu_x=\mu (N_x-M_y).
\]
Example 1.1.5:
Two C -functions u(x, y) and v(x, y) are said to be functionally dependent if
1
ux uy
det ( ) = 0, (1.1.6)
vx vy
which is a linear partial differential equation of first order for u if v is a given C -function. A large class of solutions is given
1
by
u = H (v(x, y)), (1.1.7)
where H is an arbitrary C -function.

1
Example 1.1.6: Cauchy-Riemann equations
Set f (z) = u(x, y) + iv(x, y) , where z = x + iy and u, v are given C (Ω)-functions. Here Ω is a domain in R . If the
1 2
function f (z) is differentiable with respect to the complex variable z then u, v satisfy the Cauchy-Riemann equations
ux = vy , uy = −vx . (1.1.8)
It is known from the theory of functions of one complex variable that the real part u and the imaginary part v of a differentiable
function f (z) are solutions of the Laplace equation
△u = 0, △v = 0, (1.1.9)
where △u = u xx + uyy .
Example 1.1.7: Newton Potential

The Newton potential
1
u = −−−−−−−−− − (1.1.10)
√ x2 + y 2 + z 2
is a solution of the Laplace equation in R 3

∖ (0, 0, 0) , that is, of
$$
u_{xx}+u_{yy}+u_{zz}=0.
\]

Example 1.1.8: Heat equation
Let u(x, t) be the temperature of a point x ∈ Ω at time t , where Ω ⊂ R is a domain. Then u(x, t) satisfies in Ω × [0, ∞) the
3
heat equation
ut = k△u, (1.1.11)
where △u = u x1 x1 + ux2 x2 + ux3 x3 and k is a positive constant. The condition

u(x, 0) = u0 (x), x ∈ Ω, (1.1.12)
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 ∈ ∂Ω, t ≥ 0, (1.1.13)
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 Ω
v = g(x) on ∂Ω.
Example 1.1.9: Wave equation
Figure 1.1.2: Oscillating string

The wave equation
2
utt = c △u, (1.1.14)
where u = u(x, t), c is a positive constant, describes oscillations of membranes or of three dimensional domains, for example.
In the one-dimensional case
2
utt = c uxx (1.1.15)
describes oscillations of a string.

Associated initial conditions are
u(x, 0) = u0 (x), ut (x, 0) = u1 (x), (1.1.16)
where u 0, u1 are given functions. Thus the initial position and the initial velocity are prescribed.
If the string is finite one describes additionally boundary conditions, for example
$$
u(0,t)=0,\ \ u(l,t)=0\ \ \mbox{for all}\ t\ge 0.
\]




1.2: Ordinary Differential Equations
Set
$$E(v)=\int_a^bf(x,v(x),v'(x))\ dx\]
and for given u a, ub ∈ R
$$V=\{v\in C^2[a,b]:\ v(a)=u_a,\ v(b)=u_b\},\]

where y and f is sufficiently regular. One of the basic problems in the calculus of variation is
(P) min v∈V E(v) .
Euler equation
Let u ∈ V be a solution of (P), then
$$\frac{d}{dx}f_{u'}(x,u(x),u'(x))=f_u(x,u(x),u'(x))\]
in (a, b).
Exercise 1.2.1: Proof
For fixed [a, b] with ϕ(a) = ϕ(b) = 0 and real ϵ , |ϵ| < ϵ , set g(ϵ) = E(u + ϵϕ) . Since g(0) ≤ g(ϵ)
ϕ ∈ C
2
0 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
′
Euler's equation.
Figure 1.2.1.1: Admissible Variations
Basic lemma in the calculus of variations. Let h ∈ C (a, b) and

b
∫ h(x)ϕ(x) dx = 0 (1.2.1)
a
for all ϕ ∈ C 1
0
(a, b) . Then h(x) ≡ 0 on (a, b).
Proof. Assume h(x 0) >0 for an x0 ∈ (a, b) , then there is a δ >0 such that (x0 − δ, x0 + δ) ⊂ (a, b) and h(x) ≥ h(x0 )/2 on
(x − δ, x + δ) .
0 0
Set
$$
\phi(x)
=\left\{
2
2 2
(δ − |x − x0 | ) x ∈ (x0 − δ, x0 + δ)
(1.2.2)
0 x ∈ (a, b) ∖ [ x0 − δ, x0 + δ]
\right. .
\]
Thus ϕ ∈ C 1
0
(a, b) and

$$\int_a^b h(x)\phi(x)\ dx\ge \frac{h(x_0)}{2}\int_{x_0-\delta}^{x_0+\delta}\phi(x)\ dx>0,\]
which is a contradiction to the assumption of the lemma.
□



1.3: Partial Differential Equations
The same procedure as above applied to the following multiple integral leads to a second-order quasilinear partial differential
equation. Set
$$E(v)=\int_\Omega\ F(x,v,\nabla v)\ dx,\]
where Ω ⊂ R is a domain, x = (x , … , x ), v = v(x) : Ω ↦ R , and ∇v = (v
n
1 n
1
x1 , … , vxn ) . Assume that the function F is
sufficiently regular in its arguments. For a given function h , defined on ∂Ω , set
$$V=\{v\in C^2(\overline{\Omega}):\ v=h\ \mbox{on}\ \partial\Omega\}.\]
Euler equation. Let u ∈ V be a solution of (P), then
n
∂
∑ Fux − Fu = 0 (1.3.1)
∂xi i
i=1
in Ω.
Proof. Exercise. Hint: Extend the above fundamental lemma of the calculus of variations to the case of multiple integrals. The
interval (x − δ, x + δ) in the definition of ϕ must be replaced by a ball with center at x and radius δ .
0 0 0
Example 1.2.2.1: Dirichlet integral

In two dimensions the Dirichlet integral is given by
$$D(v)=\int_\Omega\ \left(v_x^2+v_y^2\right)\ dxdy\]
and the associated Euler equation is the Laplace equation △u = 0 in Ω.
Thus, there is natural relationship between the boundary value problem
$$\triangle u=0\ \ \mbox{in}\ \Omega,\ u=h\ \ \mbox{on}\ \ \partial\Omega\]
and the variational problem
$$\min_{v\in V}\ 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, for an example see 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 finite Dirichlet integral.
The problems are equivalent, provided the given boundary value function h is in the class
(∂Ω), see Lions and Magenes [14].
1/2
H
Example 1.2.2.2: Minimal surface equation

The non-parametric minimal surface problem in two dimensions is to find a minimizer u = u(x 1, x2 ) of the problem
$$\min_{v\in V}\int_\Omega\ \sqrt{1+v_{x_1}^2+v_{x_2}^2}\ dx,\]
where for a given function h defined on the boundary of the domain Ω
$$V=\{v\in C^1(\overline{\Omega}):\ v=h\ \mbox{on}\ \partial\Omega\}.\]

Figure 1.2.2.1: Comparison surface
Suppose that the minimizer satisfies the regularity assumption u ∈ C 2
(Ω) , then
u is a solution of the minimal surface equation (Euler equation) in Ω
⎛ ⎞ ⎛ ⎞
∂ ux ∂ ux
1 2
⎜ −−−−−−−−⎟+ ⎜ −−−−−−−− ⎟ = 0. (1.3.2)

∂x1 2 ∂x2 2
⎝ √ 1 + |∇u| ⎠ ⎝ √ 1 + |∇u| ⎠
In fact, the additional assumption u ∈ C (Ω) is superfluous since it follows from regularity considerations for quasilinear
2
elliptic equations of second order, see for example Gilbarg and Trudinger [9].
Let Ω = R . Each linear function is a solution of the minimal surface equation (1.3.2). It was shown by Bernstein [2] that
2
there are no other solutions 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 define cones, see Bombieri, De Giorgi and Giusti [3].
The linearized minimal surface equation over u ≡ 0 is the Laplace equation △u = 0 . In R linear functions are solutions but
2
also many other functions in contrast to the minimal surface equation. This striking difference is caused by the strong
nonlinearity of the minimal surface equation.
More general minimal surfaces are described by using parametric representations. An example is shown in Figure 1.2.2.21. See
[18], pp. 62, for example, for rotationally symmetric minimal surfaces.
Figure 1.2.2.2: Rotationally symmetric minimal surface

1
An experiment from Beutelspacher's Mathetikum, Wissenschaftsjahr 2008, Leipzig
Neumann type boundary value problems

¯
¯¯¯
Set V =C
1
(Ω) and
$$E(v)=\int_\Omega\ F(x,v,\nabla v)\ dx-\int_{\partial\Omega}\ g(x,v)\ ds,\]
where F and g are given sufficiently regular functions and Ω ⊂ R is a bounded and sufficiently regular domain.
n
Assume u is a minimizer of E(v) in V , that is

$$u\in V:\ \ E(u)\le E(v)\ \ \mbox{for all}\ v\in V,\]

then
n
∫ ( ∑ Fu (x, u, ∇u)ϕx + Fu (x, u, ∇u)ϕ) dx

x i
i
Ω
i=1
− ∫ gu (x, u)ϕ ds = 0

∂Ω
¯
¯¯¯
for all ϕ ∈ C 1
(Ω) . Assume additionally u ∈ C 2
(Ω) , then u is a solution of the Neumann type boundary value problem
n
∂
∑ Fu − Fu = 0 in Ω
x
i
∂xi
i=1
∑ Fux νi − gu = 0 on ∂Ω,
i
i=1
where ν = (ν , … , ν ) is the exterior unit normal at the boundary

1 n ∂Ω . This follows after integration by parts from the basic
lemma of the calculus of variations.
Example 1.2.2.3: Laplace equation

Set
$$E(v)=\frac{1}{2}\int_\Omega\ |\nabla v|^2\ dx-\int_{\partial\Omega}\ h(x)v\ ds,\]
then the associated boundary value problem is
△u = 0 in Ω
∂u
= h on ∂Ω.
∂ν
Example 1.2.2.4: Capillary equation
Let Ω ⊂ R and set

2
$$E(v)=\int_\Omega\ \sqrt{1+|\nabla v|^2}\ dx+\frac{\kappa}{2}\int_\Omega\ v^2\ dx -\cos\gamma\int_{\partial\Omega}\ v\

ds.\]
Here κ is a positive constant (capillarity constant) and γ is the (constant) boundary contact angle, i. e., the angle between the
container wall and the capillary surface, defined by v = v(x , x ) , at the boundary.
1 2
Then the related boundary value problem is

div (T u) = κu in Ω
ν ⋅ T u = cos γ on ∂Ω,
where we use the abbreviation

$$Tu=\frac{\nabla u}{\sqrt{1+|\nabla u|^2}},\]
div (T u) is the left hand side of the minimal surface equation (1.3.2) and it is twice the mean curvature of the surface defined
by z = u(x , x ) , see an exercise.
1 2
The above problem describes the ascent of a liquid, water for example, in a vertical cylinder with cross section Ω. Assume the
gravity is directed downwards in the direction of the negative x -axis. Figure 1.2.2.3 shows that liquid can rise along a vertical
3
wedge which is a consequence of the strong non-linearity of the underlying equations, see Finn [7]. This photo was taken from
[15].

Figure 1.2.2.3: Ascent of liquid in a wedge



1.E: Introduction (Exercises)
These are homework exercises to accompany Miersemann's "Partial Differential Equations" Textmap. This is a textbook targeted
for a one semester first course on differential equations, aimed at engineering students. Partial differential equations are differential
equations that contains unknown multivariable functions and their partial derivatives. Prerequisite for the course is the basic
calculus sequence.
Q1.1
Find nontrivial solutions u of
$$
u_xy-u_yx=0 \ .
\]
Q1.2
Prove: In the linear space C 2 2
(R ) there are infinitely many linearly independent solutions of △u = 0 in R . 2
Hint: Real and imaginary part of holomorphic functions are solutions of the Laplace equation.
Q1.3
Find all radially symmetric functions which satisfy the Laplace equation in
R ∖ {0} for n ≥ 2 . A function u is said to be radially symmetric if u(x) = f (r), where r = (∑ .
n n 2 1/2
x )
i i
′
Hint: Show that a radially symmetric u satisfies △u = r 1−n
(r
n−1 ′
f ) by using ∇u(x) = f ′
(r)
x
r
.
Q1.4
Prove the basic lemma in the calculus of variations:
Let Ω ⊂ R be a domain and f ∈ C (Ω) such that
n
∫ f (x)h(x) dx = 0 (1.E.1)
Ω
for all h ∈ C 2
0
(Ω) . Then f ≡0 in Ω.
Q1.5
Write the minimal surface equation (1.2.2.1) as a quasilinear equation of second order.
Q1.6
Prove that a sufficiently regular minimizer in
¯
¯¯¯
C (Ω) of
1
E(v) = ∫ F (x, v, ∇v) dx − ∫ g(v, v) ds, (1.E.2)

Ω ∂Ω
is a solution of the boundary value problem

n
∂
∑ Fux − Fu = 0 in Ω
i
∂xi
i=1
∑ Fu νi − gu = 0 on ∂Ω,
x
i
i=1
where ν = (ν1 , … , νn ) is the exterior unit normal at the boundary ∂Ω .
Erich Miersemann 1.E.1 3/30/2022 https://math.libretexts.org/@go/page/2132

Q1.7
Prove that ν ⋅ T u = cos γ on ∂Ω , where γ is the angle between the container wall, which is here a cylinder, and the surface S ,
defined by z = u(x , x ) , at the boundary of S , ν is the exterior normal at
1 2
∂Ω .
Hint: The angle between two surfaces is by definition the angle between the two associated normals at the intersection of the
surfaces.
Q1.8
¯
¯¯¯
Let Ω be bounded and assume u ∈ C 2
(Ω) is a solution of
div T u = C in Ω
∇u
ν ⋅ −−−−−−−− = cos γ on ∂Ω,
2
√ 1 + |∇u|
where C is a constant.
Prove that
$$
C={|\partial\Omega|\over|\Omega|}\cos\gamma\ .
\]
Hint: Integrate the differential equation over Ω.
Q1.9
Assume Ω = 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
( − −−−−−) = κrw in 0 < r < R
√ 1 + w′2
′
w
− −−−−− = cos γ if r = R.
√ 1 + w′2
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.
Q1.10
Find all radially symmetric solutions of
′ ′
rw
( − −−−−−) = C r in 0 < r < R
√ 1 + w′2
′
w
− −−−−− = cos γ if r = R.
√ 1 + w′2
Hint: From an exercise above it follows that

$$
C=\frac{2}{R}\cos\gamma.
\]



CHAPTER OVERVIEW
2: EQUATIONS OF FIRST ORDER
The main tool for studying related problems is the theory of ordinary differential equations, which is quite different for systems of partial
differential of first order.
2.0: PRELUDE TO FIRST ORDER EQUATIONS

2.1: LINEAR EQUATIONS
2.2: QUASILINEAR EQUATIONS
2.2.1: A LINEARIZATION METHOD
2.3: NONLINEAR EQUATIONS IN TWO VARIABLES
2.4: NONLINEAR EQUATIONS IN Rn
2.5: HAMILTON-JACOBI THEORY
2.E: EQUATIONS OF FIRST ORDER (EXERCISES)
1 3/30/2022
2.0: Prelude to First Order Equations
For a given sufficiently regular function F the general equation of first order for the unknown function ) is
$$F(x,u,\nabla u)=0\]
in n . The main tool for studying related problems is the theory of ordinary differential equations. This is quite different for systems
of partial differential of first order. The general linear partial differential equation of first order can be written as
$$\sum_{i=1}^na_i(x)u_{x_i}+c(x)u=f(x)\]
for given functions ai, c and f . The general quasilinear partial differential equation of first order is
$$\sum_{i=1}^na_i(x,u)u_{x_i}+c(x,u)=0.\]



2.1: Linear Equations
Let us begin with the linear homogeneous equation
a1 (x, y)ux + a2 (x, y)uy = 0. (2.1.1)
Assume there is a C -solution z = u(x, y). This function defines a surface S which has at P
1
= (x, y, u(x, y)) the normal
1
N = −−−− −−−− (−ux , −uy , 1) (2.1.2)
2
√ 1 + |∇u|
and the tangential plane defined by
ζ − z = ux (x, y)(ξ − x) + uy (x, y)(η − y). (2.1.3)
Set p = u (x, y), q = u (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
x y
the surface element. The tangential plane is defined by the surface element. On the other hand, differential equation (2.1.1)
a1 (x, y)p + a2 (x, y)q = 0 (2.1.4)
defines at each support (x, y, z) a bundle of planes if we consider all (p, q) satisfying this equation. For fixed (x, y), this family of
planes Π(λ) = Π(λ; x, y) is defined by a one parameter family of ascents p(λ) = p(λ; x, y), q(λ) = q(λ; x, y).
The envelope of these planes is a line since
a1 (x, y)p(λ) + a2 (x, y)q(λ) = 0, (2.1.5)
which implies that the normal N(λ) on Π(λ) is perpendicular on (a 1, .

a2 , 0)
Consider a curve x(τ ) = (x(τ ), y(τ ), z(τ )) on S , let

Tx0 be the tangential plane at x = (x(τ ), y(τ ), z(τ
0 0 0 0 )) of S
and consider on T the line
x0
′ 1
L : l(σ) = x0 + σ x (τ0 ), σ ∈ R , (2.1.6)
see Figure 2.1.1.
Figure 2.1.1: Curve on a surface

We assume L coincides with the envelope, which is a line here, of the family of planes Π(λ) at . Assume that
(x, y, z)
Tx
0
= Π(λ0 ) and consider two planes
Π(λ0 ) : z − z0 = (x − x0 )p(λ0 ) + (y − y0 )q(λ0 )
Π(λ0 + h) : z − z0 = (x − x0 )p(λ0 + h) + (y − y0 )q(λ0 + h).
At the intersection l(σ) we have
(x − x0 )p(λ0 ) + (y − y0 )q(λ0 ) = (x − x0 )p(λ0 + h) + (y − y0 )q(λ0 + h). (2.1.7)

Thus,
′ ′ ′ ′
x (τ0 )p (λ0 ) + y (τ0 )q (λ0 ) = 0. (2.1.8)
From the differential equation

a1 (x(τ0 ), y(τ0 ))p(λ) + a2 (x(τ0 ), y(τ0 ))q(λ) = 0 (2.1.9)
it follows
′ ′
a1 p (λ0 ) + a2 q (λ0 ) = 0. (2.1.10)
Consequently
′
x (τ )
′ ′
(x (τ ), y (τ )) = (a1 (x(τ ), y(τ )), a2 (x(τ ), y(τ )), (2.1.11)
a1 (x(τ , y(τ ))
since τ was an arbitrary parameter. Here we assume that x (τ ) ≠ 0 and a

0
′
1 (x(τ ), y(τ )) ≠ 0 .
Then we introduce a new parameter t by the inverse of τ = τ (t) , where
τ ′
x (s)
t(τ ) = ∫ ds. (2.1.12)
τ0 a1 (x(s), y(s))
It follows x (t) = a
′
1 (x,
′
y), y (t) = a2 (x, y) . We denote x(τ (t)) by x(t) again.
Now we consider the initial value problem
′ ′
x (t) = a1 (x, y), y (t) = a2 (x, y), x(0) = x0 , y(0) = y0 . (2.1.13)
From the theory of ordinary differential equations it follows (Theorem of Picard-Lindelöf) that there is a unique solution in a
neighbourhood of t = 0 provided the functions a , a are in C . From this definition of the curves (x(t), y(t)) is follows that
1 2
1
the field of directions (a (x , y ), a (x , y )) defines the slope of these curves at (x(0), y(0)).
1 0 0 2 0 0
Definition. The differential equations in (2.1.13) are called characteristic equations or characteristic system and solutions of the
associated initial value problem are called characteristic curves.
Definition. A function ϕ(x, y) is said to be an integral of the characteristic system if ϕ(x(t), y(t)) = const. for each characteristic
curve. The constant depends on the characteristic curve considered.
Proposition 2.1. Assume ϕ ∈ C is an integral, then u = ϕ(x, y) is a solution of (2.1.1).
1
Proof. Consider for given (x 0, y0 ) the above initial value problem (2.1.13). Since ϕ(x(t), y(t)) = const. it follows
′ ′
ϕx x + ϕy y =0 (2.1.14)
for $ |t|<t_0$, t > 0 and sufficiently small.

0
Thus
$$
\phi_x(x_0,y_0)a_1(x_0,y_0)+\phi_y(x_0,y_0)a_2(x_0,y_0)=0.
\]
□
Remark. If ϕ(x, y) is a solution of equation (2.1.1) then also H (ϕ(x, y)), where H (s) is a given C -function. 1

Examples
Example 2.1.1:
Consider
a1 ux + a2 uy = 0, (2.1.15)
where a 1, a2 are constants. The system of characteristic equations is

′ ′
x = a1 , y = a2 . (2.1.16)
Thus the characteristic curves are parallel straight lines defined by

x = a1 t + A, y = a2 t + B, (2.1.17)
where A, B are arbitrary constants. From these equations it follows that

ϕ(x, y) := a2 x − a1 y (2.1.18)
is constant along each characteristic curve. Consequently, see Proposition 2.1, u = a2 x − a1 y is a solution of the differential
equation. From an exercise it follows that
u = H (a2 x − a1 y), (2.1.19)
where H (s) is an arbitrary C -function, is also a solution. Since u is constant when a x − a y is constant, equation (2.1.19)
1
2 1
defines cylinder surfaces which are generated by parallel straight lines which are parallel to the (x, y)-plane, see Figure 2.1.2.
Figure 2.1.2: Cylinder surfaces
Example 2.1.2:
Consider the differential equation
x ux + y uy = 0. (2.1.20)
The characteristic equations are

′ ′
x = x, y = y, (2.1.21)
and the characteristic curves are given by

t t
x = Ae , y = Be , (2.1.22)
where A, B are arbitrary constants. Then an integral is y/x, x ≠ 0 , and for a given C -function the function u = H (x/y) is a
1
solution of the differential equation. If y/x = const. , then u is constant. Suppose that H (s) > 0 , for example, then u defines
′
right helicoids (in German: Wendelflächen), see Figure 2.1.3.

Figure 2.1.3: Right helicoid, a 2
<x
2
+y
2
<R
2
(Museo Ideale Leonardo da Vinci, Italy)
Example 2.1.3:
y ux − x uy = 0. (2.1.23)
The associated characteristic system is

′ ′
x = y, y = −x. (2.1.24)
If follows
′ ′
x x + yy = 0, (2.1.25)
or, equivalently,
d 2 2
(x + y ) = 0, (2.1.26)
dt
which implies that x + y = const. along each characteristic. Thus, rotationally symmetric surfaces defined by
2 2
u = H (x + y ) , where H ≠ 0 , are solutions of the differential equation.

2 2 ′
Example 2.1.4:
The associated characteristic equations to
ay ux + bx uy = 0, (2.1.27)
where a, b are positive constants,

are given by
′ ′
x = ay, y = bx. (2.1.28)
It follows bx x′
− ay y
′
=0 , or equivalently,
d 2 2
(b x − ay ) = 0. (2.1.29)
dt
Solutions of the differential equation are u = H (bx − ay ) , which define surfaces which have a hyperbola as the intersection
2 2
with planes parallel to the (x, y)-plane. Here H (s) is an arbitrary C -function, H (s) ≠ 0 . 1 ′



2.2: Quasilinear Equations
Here we consider the equation
a1 (x, y, u)ux + a2 (x, y, u)uy = a3 (x, y, u). (2.2.1)
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 (2.2.1).
One arrives at characteristic equations x = a , y = a , z = a from (2.2.1) by the same arguments as in the case of
′
1
′
2
′
3
homogeneous linear equations in two variables. The additional equation 3 follows from
′ ′ ′
z (τ ) = p(λ)x (τ ) + q(λ)y (τ )
= p a1 + q a2
= a3 ,
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 (2.2.1) into a homogeneous linear equation for an unknown function of three
variables by the following trick.
We are looking for a function ψ(x, y, u) such that the solution u = u(x, y) of Equation (2.2.1) is defined implicitly by
ψ(x, y, u) = const. Assume there is such a function ψ and let u be a solution of (2.2.1), then
ψx + ψu ux = 0, ψy + ψu uy = 0. (2.2.1.1)
Assume ψ u ≠0 , then
ψx ψy
ux = − , uy = − . (2.2.1.2)
ψu ψu
From (2.2.1) we obtain

a1 (x, y, z)ψx + a2 (x, y, z)ψy + a3 (x, y, z)ψz = 0, (2.2.1.1)
where z := u .
We consider the associated system of characteristic equations
′
x (t) = a1 (x, y, z)
′
y (t) = a2 (x, y, z)
′
z (t) = a3 (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 , w = w(x, y, z), is an integral, i. e., it is constant along each fixed solution of (2.2.1.1), then
1
ψ = w(x, y, z) is a solution of (2.2.1.1).
(ii) The function z = u(x, y), implicitly defined through ψ(x, u, z) = const. , is a solution of (2.2.1), provided that ψz ≠0 .
(iii) Let z = u(x, y) be a solution of (2.2.1) and let (x(t), y(t)) be a solution
of
′ ′
x (t) = a1 (x, y, u(x, y)), y (t) = a2 (x, y, u(x, y)), (2.2.1.3)
then z(t) := u(x(t), y(t)) satisfies the third of the above characteristic equations.
Proof. Exercise.


Erich Miersemann 2.2.1.1 3/30/2022 https://math.libretexts.org/@go/page/2168

2.2.2: Initial Value Problem of Cauchy
Consider again the quasilinear equation
(⋆) a 1 (x, y, u)ux + a2 (x, y, u)uy = a3 (x, y, u) .
Let
Γ : x = x0 (s), y = y0 (s), z = z0 (s), s1 ≤ s ≤ s2 , − ∞ < s1 < s2 < +∞ (2.2.2.1)
be a regular curve in R and denote by C the orthogonal projection of Γ onto the (x, y)-plane, i. e.,
3
$$
\mathcal{C}:\ \ x=x_0(s),\ \ y=y_0(s).
\]
Initial value problem of Cauchy: Find a C -solution u = u(x, y) of
1
(⋆) such that u(x0 (s), y0 (s)) = z0 (s) , i. e., we seek a
surface S defined by z = u(x, y) which contains the curve Γ.
Figure 2.2.2.1: Cauchy initial value problem

Definition. The curve Γ is said to be non-characteristic if
$$
x_0'(s)a_2(x_0(s),y_0(s))-y_0'(s)a_1(x_0(s),y_0(s))\not=0.
\]
Theorem 2.1. Assume a 1, a2 , a2 ∈ C
1
in their arguments, the initial data x 0, y0 , z0 ∈ C
1
[ s1 , s2 ] and Γ is non-characteristic.
Then there is a neighborhood of C 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) = a1 (x, y, z)
′
y (t) = a2 (x, y, z)
′
z (t) = a3 (x, y, z)
with the initial conditions
x(s, 0) = x0 (s)
y(s, 0) = y0 (s)
z(s, 0) = z0 (s).
Let x = x(s, t) , y = y(s, t) , z = z(s, t) be the solution, s ≤ s ≤ s , |t| < η for an η > 0 . We will show that this set of curves,
1 2
see Figure 2.2.2.1, defines 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
neighborhood of t = 0 since
∂(x, y) ∣ xs xt ∣
∣ ′ ′
det =∣ ∣ = x (s)a2 − y (s)a1 ≠ 0, (2.2.2.2)
∣ 0 0
∂(s, t) t=0 ∣ ys yt ∣
t=0

and the initial curve Γ is non-characteristic by assumption.
Set
u(x, y) := z(s(x, y), t(x, y)), (2.2.2.3)
then u satisfies the initial condition since
u(x, y)|t=0 = z(s, 0) = z0 (s). (2.2.2.4)
The following calculation shows that u is also a solution of the differential equation (⋆).
a1 ux + a2 uy = a1 (zs sx + zt tx ) + a2 (zs sy + zt ty )
= zs (a1 sx + a2 sy ) + zt (a1 tx + a2 ty )
= zs (sx xt + sy yt ) + zt (tx xt + ty yt )
= a3
since 0 = s t = sx xt + sy yt and 1 = t
t = tx xt + ty yt .
(ii) Uniqueness. Suppose that v(x, y) is a second solution. Consider a point (x , y ) in a neighborhood of the curve (x
′ ′
0 (s), y(s)) ,
s − ϵ ≤ s ≤ s + ϵ , ϵ > 0 small. The inverse parameters are s = s(x , y ) , t = t(x , y ) , see Figure 2.2.2.2.
′ ′ ′ ′ ′ ′
1 2
Figure 2.2.2.2: Uniqueness proof

Let
′ ′ ′
A : x(t) := x(s , t), y(t) := y(s , t), z(t) := z(s , t) (2.2.2.5)
be the solution of the above initial value problem for the characteristic differential equations with the initial data
′ ′ ′ ′ ′ ′
x(s , 0) = x0 (s ), y(s , 0) = y0 (s ), z(s , 0) = z0 (s ). (2.2.2.6)
According to its construction this curve is on the surface S defined by u = u(x, y) and u(x , y ′ ′ ′ ′
) = z(s , t ) . Set
ψ(t) := v(x(t), y(t)) − z(t), (2.2.2.7)
then
′ ′ ′ ′
ψ (t) = vx x + vy y − z
= xx a1 + vy a2 − a3 = 0
and
′ ′ ′
ψ(0) = v(x(s , 0), y(s , 0)) − z(s , 0) = 0 (2.2.2.8)
since v is a solution of the differential equation and satisfies the initial condition by assumption. Thus, ψ(t) ≡ 0 , i. e.,

′ ′ ′
v(x(s , t), y(s , t)) − z(s , t) = 0. (2.2.2.9)
Set t = t , then
′
′ ′ ′ ′
v(x , y ) − z(s , t ) = 0, (2.2.2.10)
which shows that v(x , y ′ ′ ′ ′

) = u(x , y ) because of z(s , t ) = u(x , y ) .
′ ′ ′ ′
Remark. In general, there is no uniqueness if the initial curve Γ is a characteristic curve, see an exercise and Figure 2.2.2.3, which
illustrates this case.
Figure 2.2.2.3: Multiple solutions
Examples
Example 2.2.2.1:
Consider the Cauchy initial value problem

ux + uy = 0 (2.2.2.11)
with the initial data

1
x0 (s) = s, y0 (s) = 1, z0 (s) is a given C -function. (2.2.2.12)
These initial data are non-characteristic since ′

y a1 − x a2 = −1
0
′
0
. The solution of the associated system of characteristic
equations
′ ′ ′
x (t) = 1, y (t) = 1, u (t) = 0 (2.2.2.13)
with the initial conditions
x(s, 0) = x0 (s), y(s, 0) = y0 (s), z(s, 0) = z0 (s) (2.2.2.14)
is given by
x = t + x0 (s), y = t + y0 (s), z = z0 (s), (2.2.2.15)
i. e.,
x = t + s, y = t + 1, z = z0 (s). (2.2.2.16)
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.

Example 2.2.2.2:
A problem from kinetics in chemistry. Consider for x ≥ 0 , y ≥ 0 the problem
−k1 x
ux + uy = (k0 e + k2 ) (1 − u) (2.2.2.17)
with initial data
u(x, 0) = 0, x > 0, and u(0, y) = u0 (y), y > 0. (2.2.2.18)
Here the constants k are positive, these constants define the velocity of the reactions in consideration, and the function u (y)
j 0
is given. The variable x is the time and y is the height 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 C1 ∪ C2 of the initial curve onto the
(x, y)-plane has a corner at the origin, see Figure 2.2.2.4.
Figure 2.2.2.4: Domains to the chemical kinetics example

The associated system of characteristic equations is
′ ′ ′ −k1 x
x (t) = 1, y (t) = 1, z (t) = (k0 e + k2 ) (1 − z). (2.2.2.19)
It follows x = t + c , y = t + c with constants c . Thus the projection of the characteristic curves on the (x, y)-plane are
1 2 j
straight lines parallel to y = x . We will solve the initial value problems in the domains Ω and Ω , see Figure 2.2.2.4, 1 2
separately.
(i) The initial value problem in Ω . The initial data are
1
x0 (s) = s, y0 (s) = 0, z0 (0) = 0, s ≥ 0. (2.2.2.20)
It follows
x = x(s, t) = t + s, y = y(s, t) = t. (2.2.2.21)
Thus
′ −k1 (t+s)
z (t) = (k0 e + k2 )(1 − z), z(0) = 0. (2.2.2.22)
The solution of this initial value problem is given by

k0 k0
−k1 (s+t) −k1 s
z(s, t) = 1 − exp( e − k2 t − e ). (2.2.2.23)
k1 k1
Consequently
k0
−k1 x −k1 (x−y)
u1 (x, y) = 1 − exp( e − k2 y − k0 k1 e ) (2.2.2.24)
k1

is the solution of the Cauchy initial value problem in Ω . If time x tends to ∞, we get the limit
1
$$
\lim_{x\to\infty} u_1(x,y)=1-e^{-k_2y}.
\]
(ii) The initial value problem in Ω . The initial data are here
2
x0 (s) = 0, y0 (s) = s, z0 (0) = u0 (s), s ≥ 0. (2.2.2.25)
It follows
x = x(s, t) = t, y = y(s, t) = t + s. (2.2.2.26)
Thus
′ −k1 t
z (t) = (k0 e + k2 )(1 − z), z(0) = 0. (2.2.2.27)
The solution of this initial value problem is given by

k0 k0
−k1 t
z(s, t) = 1 − (1 − u0 (s)) exp( e − k2 t − ). (2.2.2.28)
k1 k1
Consequently
k0 k0
−k1 x
u2 (x, y) = 1 − (1 − u0 (y − x)) exp( e − k2 x − ) (2.2.2.29)
k1 k1
is the solution in Ω .
2
If x = y , then
k0 k0
−k1 x
u1 (x, y) = 1 − exp( e − k2 x − )
k1 k1
k0 k0
−k1 x
u2 (x, y) = 1 − (1 − u0 (0)) exp( e − k2 x − ).
k1 k1
If u (0) > 0 , then u

0 1 < u2 if x = y , i. e., there is a jump of the concentration of the substrate along its burning front defined
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
a1 (x, y)ux + a2 (x, y)uy = 0 (2.2.2.30)
is known.
Let
x0 (s), y0 (s), z0 (s), s1 < s < s2 (2.2.2.31)

be the initial data and let u = ϕ(x, y) be a solution of the differential equation. We assume that
′ ′
ϕx (x0 (s), y0 (s))x (s) + ϕy (x0 (s), y0 (s))y (s) ≠ 0 (2.2.2.32)
0 0
is satisfied. Set
g(s) = ϕ(x0 (s), y0 (s)) (2.2.2.33)
and let s = h(g) be the inverse function.

The solution of the Cauchy initial problem is given by u 0 (h(ϕ(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\left(h(\phi(x_0(s),y_0(s))\right)=u_0(h(g))=u_0(s).
\]
Example 2.2.2.3:
Consider equation
ux + uy = 0 (2.2.2.34)
with initial data

x0 (s) = s, y0 (s) = 1, u0 (s) is a given function. (2.2.2.35)
A solution of the differential equation is ϕ(x, y) = x − y . Thus

ϕ((x0 (s), y0 (s)) = s − 1 (2.2.2.36)
and
u0 (ϕ + 1) = u0 (x − y + 1) (2.2.2.37)
is the solution of the problem.



2.3: Nonlinear Equations in Two Variables
Here we consider equation
F (x, y, z, p, q) = 0, (2.3.1)
where z = u(x, y), p = u x (x, y) ,q=u y (x, y) and F ∈ C

2
is given such that F p
2
+ Fq
2
≠0 .
In contrast to the quasilinear case, this general nonlinear equation is more complicated.
Together with (2.3.1) 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 (2.3.1).
′
x (t) = Fp (2.3.2)
′
y (t) = Fq (2.3.3)
′
z (t) = p Fp + q Fq (2.3.4)
′
p (t) = −Fx − Fu p (2.3.5)
′
q (t) = −Fy − Fu q. (2.3.6)
Definition. Equations (2.3.2)--(2.3.6) are said to be characteristic equations of equation (2.3.1) 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 2.3.1: Gaspard Monge (Panthèon, Paris)

We will see, as in the quasilinear case, that the strips defined by the characteristic equations build the solution surface of the
Cauchy initial value problem.
Let z = u(x, y) be a solution of the general nonlinear differential equation (2.3.1).
Let (x 0, be fixed, then equation (2.3.1) defines a set of planes given by
y0 , z0 )
, i. e., planes given by z = v(x, y) which contain the point (x , y , z )

(x0 , y0 , z0 , p, q) 0 0 0
and for which v = p , v = q at (x , y ). In the case of quasilinear equations these set of planes is a bundle of planes which all
x y 0 0
contain a fixed straight line, see Section 2.1. In the general case of this section the situation is more complicated.
Consider the example
2 2
p +q = f (x, y, z), (2.3.7)
where f is a given positive function. Let E be a plane defined by z = v(x, y) and which contains (x 0, y0 , z0 ) . Then the normal on
the plane E directed downward is
$${\bf N}=\frac{1}{\sqrt{1+|\nabla v|^2}}(p,q,-1),\]
where p = v (x , y ) , q = v (x , y ) . It follows from (2.3.7) that the normal N makes a constant angle with the z -axis, and the
x 0 0 y 0 0
z -coordinate of N is constant, see Figure 2.3,2.

Figure 2.3.2: Monge cone in an example
Thus the endpoints of the normals fixed at (x 0, y0 , z0 ) define 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 (2.3.1) defines 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 2.3.3.
Figure 2.3.3: Monge cones

More precisely, we assume there exists, as in the above example, a one parameter C -family 1
$$p(\lambda)=p(\lambda;x,y,z),\ q(\lambda)=q(\lambda;x,y,z)\]
of solutions of (2.3.1). These (p(λ), q(λ)) define a family Π(λ) of planes.
Let
$${\bf x}(\tau)=(x(\tau),y(\tau),z(\tau))\]
be a curve on the surface S which touches at each point its Monge cone, see Figure 2.3.4. Thus we assume that at each point of the
surface S the associated tangent plane coincides with a plane from the family Π(λ) at this point.
Figure 2.3.4: Monge cones along a curve on the surface

Consider the tangential plane T x0 of the surface S at x 0 = (x(τ0 ), y(τ0 ), z(τ0 )) . The straight line
$${\bf l}(\sigma)={\bf x}_0+\sigma {\bf x}'(\tau_0),\qquad -\infty<\sigma<\infty,\]
is an apothem (in German: Mantellinie) of the cone by assumption and is contained in the tangential plane T x0 as the tangent of a
curve on the surface S . It is defined through
′ ′
x (τ0 ) = l (σ). (2.3.8)
The straight line l(σ) satisfies

$$l_3(\sigma)-z_0=(l_1(\sigma)-x_0)p(\lambda_0)+(l_2(\sigma)-y_0)q(\lambda_0),\]

since it is contained in the tangential plane T x0 defined by the slope (p, q). It follows
$$l_3'(\sigma)=p(\lambda_0)l_1'(\sigma)+q(\lambda_0)l_2'(\sigma).\]
Together with (2.3.8) we obtain
′ ′ ′
z (τ ) = p(λ0 )x (τ ) + q(λ0 )y (τ ). (2.3.9)
The above straight line l is the limit of the intersection line of two neighboring planes which envelopes the Monge cone:
z − z0 = (x − x0 )p(λ0 ) + (y − y0 )q(λ0 )
z − z0 = (x − x0 )p(λ0 + h) + (y − y0 )q(λ0 + h).
On the intersection one has

$$(x-x_0)p(\lambda)+(y-y_0)q(\lambda_0)=(x-x_0)p(\lambda_0+h)+(y-y_0)q(\lambda_0+h).\]
Let h → 0 , it follows
$$(x-x_0)p'(\lambda_0)+(y-y_0)q'(\lambda_0)=0.\]
Since x = l 1 (σ) ,y =l 2 (σ) in this limit position, we have
$$p'(\lambda_0)l_1'(\sigma)+q'(\lambda_0)l_2'(\sigma)=0,\]
and it follows from (2.3.8) that
′ ′ ′ ′
p (λ0 )x (τ ) + q (λ0 )y (τ ) = 0. (2.3.10)
From the differential equation F (x 0, y0 , z0 , p(λ), q(λ)) = 0 we see that

′ ′
Fp p (λ) + Fq q (λ) = 0. (2.3.11)
Assume x (τ
′
0) ≠0 and Fp ≠0 , then we obtain from (2.3.10), (2.3.11)
′
y (τ0 ) Fq
= ,
′
x (τ0 ) Fp
and from (2.3.9) (2.3.11) that

′
z (τ0 ) Fq
= p +q .
x′ (τ0 ) Fp
It follows, since τ was an arbitrary fixed parameter,

0
′ ′ ′ ′
x (τ ) = (x (τ ), y (τ ), z (τ ))
Fq Fq
′ ′ ′
= ( x (τ ), x (τ ) , x (τ ) (p + q ))
Fp Fp
′
x (τ )
= (Fp , Fq , p Fp + q Fq ),
Fp
i. e., the tangential vector x (τ ) is proportional to (F

′
p, Fq , p Fp + q Fq ) .
Set
′
x (τ )
a(τ ) = , (2.3.12)
Fp
where F = F (x(τ ), y(τ ), z(τ ), p(λ(τ )), q(λ(τ ))) . Introducing the new parameter t by the inverse of τ = τ (t) , where
τ
t(τ ) = ∫ a(s) ds, (2.3.13)

τ0

we obtain the characteristic equations (2.3.2)--(2.3.4).
Here we denote x(τ (t)) by x(t) again.
From the differential equation (2.3.1) and from (2.3.2)--(2.3.4) we get equations (2.3.5) and (2.3.6). Assume the surface
z = u(x, y) under consideration is in C , then
2
Fx + Fz p + Fp px + Fq py = 0, (qx = py )
′ ′
Fx + Fz p + x (t)px + y (t)py = 0
′
Fx + Fz p + p (t) = 0
since p = p(x, y) = p(x(t), y(t)) on the curve x(t) . Thus equation (2.3.5) of the characteristic system is shown. Differentiating
the differential equation (2.3.1) with respect to y , we get finally equation (2.3.6).
Remark. In the previous quasilinear case
F (x, y, z, p, q) = a1 (x, y, z)p + a2 (x, y, z)q − a3 (x, y, z) (2.3.14)
the first three characteristic equations are the same:

′ ′ ′
x (t) = a1 (x, y, z), y (t) = a2 (x, y, z), z (t) = a3 (x, y, z). (2.3.15)
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 non-characteristic. That is, we do not need the other remaining two
characteristic equations.
The other two equations (2.3.5) and (2.3.6) are satisfied in this quasilinear case automatically if there is a solution of the equation,
see the above derivation of these equations.
The geometric meaning of the first three characteristic differential equations (2.3.2)--(2.3.6) is the following one. Each point of the
curve
A : (x(t), y(t), z(t)) (2.3.16)
corresponds a tangential plane with the normal direction (−p, −q, 1) such that
′ ′ ′
z (t) = p(t)x (t) + q(t)y (t). (2.3.17)
This equation is called strip condition. On the other hand, let z = u(x, y) defines a surface, then z(t) := u(x(t), y(t)) satisfies the
strip condition, where p = u and q = u , that is, the ''scales'' defined by the normals fit together.
x y
Proposition 2.3. F (x, y, z, p, q) is an integral, i. e., it is constant along each characteristic curve.
Proof.
d
′ ′ ′ ′ ′
F (x(t), y(t), z(t), p(t), q(t)) = Fx x + Fy y + Fz z + Fp p + Fq q
dt
= Fx Fp + Fy Fq + p Fz Fp + q Fz Fq
−Fp fx − Fp Fz p − Fq Fy − Fq Fz q
= 0.
Corollary. Assume F (x 0, y0 , z0 , p0 , q0 ) = 0 , then F =0 along characteristic curves with the initial data (x 0, y0 , z0 , p0 , q0 ) .
Proposition 2.4.
Let z = u(x, y), u ∈ C , be a solution of the nonlinear equation (2.3.1). Set
2
z0 = u(x0 , y0 , ) p0 = ux (x0 , y0 ), q0 = uy (x0 , y0 ). (2.3.18)

Then the associated characteristic strip is in the surface S , defined by z = u(x, y).
Thus
z(t) = u(x(t), y(t))
p(t) = ux (x(t), y(t))
q(t) = uy (x(t), y(t)),
where (x(t), y(t), z(t), p(t), q(t)) is the solution of the characteristic system (2.3.2)--(2.3.6) with initial data (x 0, y0 , z0 , p0 , q0 )
Proof. Consider the initial value problem

′
x (t) = Fp (x, y, u(x, y), ux (x, y), uy (x, y))
′
y (t) = Fq (x, y, u(x, y), ux (x, y), uy (x, y))
with the initial data x(0) = x , y(0) = y . We will show that

0 0
(x(t), y(t), u(x(t), y(t)), ux (x(t), y(t)), uy (x(t), y(t))) (2.3.19)
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)) ⊂ S , and
′ ′ ′
z (t) = ux x (t) + uy y (t) = ux Fp + uy Fq . (2.3.20)
Set p(t) = u x (x(t), y(t)), q(t) = uy (x(t), y(t)) , then

′
p (t) = uxx Fp + uxy Fq
′
q (t) = uyx Fp + uyy Fq .
Finally, from the differential equation F (x, y, u(x, y), u x (x, it follows
y), uy (x, y)) = 0
′
p (t) = −Fx − Fu p
′
q (t) = −Fy − Fu q.



2.3.1: Initial Value Problem of Cauchy
Let
x = x0 (s), y = y0 (s), z = z0 (s), p = p0 (s), q = q0 (s), s1 < s < s2 , (2.3.1.1)
be a given initial strip such that the strip condition

′ ′ ′
z (s) = p0 (s)x (s) + q0 (s)y (s) (2.3.1.2)
0 0 0
is satisfied. Moreover, we assume that the initial strip satisfies the nonlinear equation, that is,
F (x0 (s), y0 (s), z0 (s), p0 (s), q0 (s)) = 0. (2.3.1.3)
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 S defined by
z = u(x, y) contains the above initial strip.
Similar to the quasilinear case we will show that the set of strips defined by the characteristic system which are fixed onto the
initial strip, see Figure 2.3.1.1, fit together and define the surface for which we are looking at.
Figure 2.3.1.1: Construction of the solution

Definition. A strip (x(τ ), y(τ ), z(τ ), p(τ ), q(τ )), τ 1 < τ < τ2 , is said to be non-characteristic if
$$x'(\tau)F_q(x(\tau),y(\tau),z(\tau),p(\tau),q(\tau))-y'(\tau)F_p(x(\tau),y(\tau),z(\tau),p(\tau),q(\tau))\not=0.\]
Theorem 2.2. For a given non-characteristic initial strip (2.3.1.1), x , y , z ∈ C and p , q ∈ C which satisfies the strip
0 0 0
2
0 0
1
condition (2.3.1.2) and the differential equation (2.3.1.3) there exists exactly one solution z = u(x, y) of the Cauchy initial value
problem in a neighborhood of the initial curve (x (s), y (s), z (s)) , i. e., z = u(x, y) is the solution of the differential equation
0 0 0
(2.3.1) and
u(x0 (s), y0 (s)) = z0 (s)
ux (x0 (s), y0 (s)) = p0 (s)
uy (x0 (s), y0 (s)) = q0 (s).
Proof. Consider the system (2.7)--(2.11) with initial data

x(s, 0) = x0 (s), y(s, 0) = y0 (s), z(s, 0) = z0 (s)
p(s, 0) = p0 (s), q(s, 0) = q0 (s).
We will show that the surface defined by x = x(s, t), y(s, t) is the surface defined by z = u(x, y), where u is the solution of the
Cauchy initial value 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 neighborhood of t = 0 . This inverse exists since the initial strip is non-characteristic by assumption:
$$\det\frac{\partial(x,y)}{\partial(s,t)}\Big|_{t=0}=x_0F_q -y_0F_q\not=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) = ux (x, y) and
Q(x, y) = u (x, y). To see this, we consider the function
y
$$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 satisfies the strip condition by assumption. In the following we will find that for fixed s the function h
satisfies a linear homogeneous ordinary differential equation of first order. Consequently, h(s, t) = 0 in a neighborhood of t = 0 .
Thus the strip condition is also satisfied along transversal strips to the characteristic strips, see Figure 2.3.1.1.
Then the set of ''scales'' fit together and define a surface like the scales of a fish.
From the definition of h(s, t) and the characteristic equations we get
ht (s, t) = zst − pt xs − qt ys − p xst − q yst
∂
= (zt − p xt − q yt ) + ps xt + qs yt − qt ys − pt xs
∂s
= (p xs + q ys )Fz + Fx xs + Fy zs + Fp ps + Fq qs .
Since F (x(s, t), y(s, t), z(s, t), p(s, t), q(s, t)) = 0 , it follows after differentiation of this equation with respect to s the
differential equation
$$h_t=-F_zh.\]
Hence h(s, t) ≡ 0 , since h(s, 0) = 0 .
Thus we have
zs = p xs + q ys
zt = p xt + q yt
zs = ux xs + uy ys
zt = ux yt + uy yt .
The first 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 − ux )xs + (Q − uy )ys = 0
(P − ux )xt + (Q − uy )yt = 0.
It follows P = ux and Q = u .
y
The initial conditions

u(x(s, 0), y(s, 0)) = z0 (s)
ux (x(s, 0), y(s, 0)) = p0 (s)
uy (x(s, 0), y(s, 0)) = q0 (s)
are satisfied since

u(x(s, t), y(s, t)) = z(s(x, y), t(x, y)) = z(s, t)
ux (x(s, t), y(s, t)) = p(s(x, y), t(x, y)) = p(s, t)
uy (x(s, t), y(s, t)) = q(s(x, y), t(x, y)) = q(s, t).
The uniqueness follows as in the proof of Theorem 2.1.

□
Example 2.3.1.1:
A differential equation which occurs in the geometrical optic is

$$u_x^2+u_y^2=f(x,y),\]
where the positive function f (x, y) is the index of refraction. The level sets defined by u(x, y) = const. are called {\it 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 the equation is
3
$$u_x^2+u_y^2+u_z^2=f(x,y,z).\]
Thus we have to extend the previous theory from R to R , n ≥ 3 .
2 n



2.4: Nonlinear Equations in R n Rn
Here we consider the nonlinear differential equation
F (x, z, p) = 0, (2.4.1)
where
n 1
x = (x1 , … , xn ), z = u(x) : Ω ⊂ R ↦ R , p = ∇u. (2.4.2)
The following system of 2n + 1 ordinary differential equations is called characteristic system.

′
x (t) = ∇p F
′
z (t) = p ⋅ ∇p F
′
p (t) = −∇x F − Fz p.
Let
x0 (s) = (x01 (s), … , x0n (s)), s = (s1 , … , sn−1 ), (2.4.3)
be a given regular (n-1)-dimensional C -hypersurface in R , i. e., we assume

2 n
∂ x0 (s)
rank = n − 1. (2.4.4)
∂s
Here s ∈ D is a parameter from an (n − 1) -dimensional parameter domain D.

For example, x = x 0 (s) defines in the three dimensional case a regular surface in R . 3
Assume
1
z0 (s) : D ↦ R , p0 (s) = (p01 (s), … , p0n (s)) (2.4.5)
are given sufficiently regular functions.

The (2n + 1) -vector
(x0 (s), z0 (s), p0 (s)) (2.4.6)
is called initial strip manifold and the condition

n−1
∂z0 ∂x0i
= ∑ p0i (s) , (2.4.7)
∂sl ∂sl
i=1
l = 1, … , n − 1 , strip condition.
The initial strip manifold is said to be non-characteristic if
Fp Fp ⋯ Fp
⎛ 1 2 n ⎞
∂x ∂x02 ∂x0n
⎜ 01
⋯ ⎟
⎜ ∂s ∂s1 ∂s1
⎟
det ⎜ ⎟ ≠ 0,
1
(2.4.8)
⎜ ⎟
⎜... ... ... ... ⎟
⎜ ⎟
∂x01 ∂x02 ∂x0n
⎝ ⋯ ⎠
∂sn−1 ∂sn−1 ∂sn−1

where the argument of F is the initial strip manifold.
pj
Initial value problem of Cauchy. Seek a solution z = u(x) of the differential equation (2.4.1) such that the initial manifold is a
subset of {(x, u(x), ∇u(x)) : x ∈ Ω}.
As in the two dimensional case we have under additional regularity assumptions
Theorem 2.3. Suppose the initial strip manifold is not characteristic and satisfies differential equation (2.4.1), that is,
F (x (s), z (s), p (s)) = 0 . Then there is a neighborhood of the initial manifold (x (s), z (s)) such that there exists a unique
0 0 0 0 0
solution of the Cauchy initial value problem.

Sketch of proof. Let
x = x(s, t), z = z(s, t), p = p(s, t) (2.4.9)
be the solution of the characteristic system and let

s = s(x), t = t(x) (2.4.10)
be the inverse of x = x(s, t) which exists in a neighborhood of t = 0 . Then, it turns out that
z = u(x) := z(s1 (x1 , … , xn ), … , sn−1 (x1 , … , xn ), t(x1 , … , xn )) (2.4.11)
is the solution of the problem.



2.5: Hamilton-Jacobi Theory
The nonlinear equation (2.4.1) of previous section in one more dimension is
$$
F(x_1,\ldots,x_n,x_{n+1},z,p_1,\ldots,p_n,p_{n+1})=0.
\]
The content of the Hamilton1-Jacobi2 theory is the theory of the special case
F ≡ pn+1 + H (x1 , … , xn , xn+1 , p1 , … , pn ) = 0, (2.5.1)
i. e., the equation is linear in p n+1 and does not depend on z explicitly.
Remark. Formally, one can write equation (2.4.1)
$$F(x_1,\ldots,x_n,u,u_{x_1},\ldots,u_{x_n})=0\]
as an equation of type (2.5.1). Set x n+1 =u and seek u implicitly from
$$\phi(x_1,\ldots,x_n,x_{n+1})=const.,\]
where ϕ is a function which is defined by a differential equation.
Assume ϕ xn+1 ≠0 , then
0 = F (x1 , … , xn , u, ux1 , … , uxn )
ϕx1 ϕxn
= F (x1 , … , xn , xn+1 , − ,…,− )
ϕxn+1 ϕxn+1
= : G(x1 , … , xn+1 , ϕ1 , … , ϕx ).
n+1
Suppose that G ϕx
n+1
≠0 , then
$$\phi_{x_{n+1}}=H(x_1,\ldots,x_n,x_{n+1},\phi_{x_1},\ldots,\phi_{x_{n+1}}).\]
The associated characteristic equations to (2.5.1) are
′
x (τ ) = Fp =1
n+1 n+1
′
x (τ ) = Fp = Hp , k = 1, … , n
k k k
n+1 n
′
z (τ ) = ∑ pl Fp = ∑ pl Hp + pn+1
l l
l=1 l=1
= ∑ pl Hp − H
l
l=1
′
p (τ ) = −Fx − Fz pn+1
n+1 n+1
= −Fxn+1
′
p (τ ) = −Fx − Fz pk
k k
= −Fx , k = 1, … , n.
k
Set t := x n+1 , then we can write partial differential equation (2.5.1) as

ut + H (x, t, ∇x u) = 0 (2.5.2)
and 2n of the characteristic equations are

′
x (t) = ∇p H (x, t, p) (2.5.3)
′
p (t) = −∇x H (x, t, p). (2.5.4)
Here is
$$x=(x_1,\ldots,x_n),\ p=(p_1,\ldots,p_n).\]

Let x(t), p(t) be a solution of (2.5.3) and (2.5.4), then it follows p ′
n+1
(t) and z ′
(t) from the characteristic equations
′
p (t) = −Ht
n+1
′
z (t) = p ⋅ ∇p H − H .
Definition. The function H (x, t, p) is called Hamilton function, equation (2.5.1) Hamilton-Jacobi equation and the system (2.5.3),
(2.5.4) 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 construct 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 differential equations.
Definition.
A solution ϕ(a; x, t) of the Hamilton-Jacobi equation, where a = (a1 , … , an ) is an n -tuple of real parameters, is called a
complete integral of the Hamilton-Jacobi equation if
$$
\det (\phi_{x_ia_l})_{i,l=1}^n\not=0.
\]
Remark. If u is a solution of the Hamilton-Jacobi equation, then also u + const.
Theorem 2.4 (Jacobi). Assume
2
u = ϕ(a; x, t) + c, c = const. , ϕ ∈ C in its arguments, (2.5.5)
is a complete integral. Then one obtains by solving of
bi = ϕai (a; x, t) (2.5.6)
with respect to x l = xl (a, b, t) , where b i i = 1, … , n are given real constants, and then by setting
pk = ϕxk (a; x(a, b; t), t) (2.5.7)
a 2n-parameter family of solutions of the canonical system.

Proof. Let
$$x_l(a,b;t),\ l=1,\ldots,n,\]
be the solution of the above system. The solution exists since ϕ is a complete integral by assumption. Set
$$p_k(a,b;t)=\phi_{x_k}(a;x(a,b;t),t),\ k=1,\ldots,n.\]
We will show that x and p solves the canonical system. Differentiating ϕ ai = bi with respect to t and the Hamilton-Jacobi equation
ϕ + H (x, t, ∇ ϕ) = 0 with respect to a , we obtain for i = 1, … , n
t x i
n
∂xk
ϕtai + ∑ ϕxk ai = 0
∂t
k=1
ϕtai + ∑ ϕxk ai H pk = 0.
k=1
Since ϕ is a complete integral it follows for k = 1, … , n

$$\frac{\partial x_k}{\partial t}=H_{p_k}.\]
∂xk
Along a trajectory, i. e., where a, b are fixed, it is ∂t
= x (t)
′
k
. Thus
$$x_k'(t)=H_{p_k}.\]
Now we differentiate p i (a, b; t) with respect to t and ϕ t + H (x, t, ∇x ϕ) = 0 with respect to x , and obtain
i

n
′ ′
p (t) = ϕxi t + ∑ ϕxi xk x (t)
i k
k=1
0 = ϕx t + ∑ ϕx xk Hp + Hx
i i k i
k=1
′
0 = ϕx t + ∑ ϕx xk x (t) + Hx
i i k i
k=1
It follows finally that p

′
i
(t) = −Hx
i
.
□
Example 2.5.1: Kepler problem

The motion of a mass point in a central field takes place in a plane, say the (x, y) -plane, see Figure 2.5.1, and satisfies the
system of ordinary differential equations of second order
$$x''(t)=U_x,\ y''(t)=U_y,\]
where
$$U(x,y)=\frac{k^2}{\sqrt{x^2+y^2}}.\]
Here we assume that k is a positive constant and that the mass point is attracted of the origin. In the case that it is pushed one
2
has to replace U by −U . See Landau and Lifschitz [12], Vol 1, for instance, concerning the related physics.
Figure 2.5.1: Motion in a central field

Set
$$p=x',\ q=y'\]
and
$$H=\frac{1}{2}(p^2+q^2)-U(x,y),\]
then
′ ′
x (t) = Hp , y (t) = Hq
′ ′
p (t) = −Hx , q (t) = −Hy .
The associated Hamilton-Jacobi equation is

2
1 k
2 2
ϕt + (ϕx + ϕy ) = − −−−− −.
2 √ x2 + y 2
which is in polar coordinates (r, θ)

2
1 1 k
2 2
ϕt + (ϕr + ϕ ) = . (2.5.8)
2 θ
2 r r
Now we will seek a complete integral of (2.5.8) by making the ansatz
ϕt = −α = const. ϕθ = −β = const. (2.5.9)

and obtain from (2.5.8) that
−−−−−−−−−−−−−
r 2 2
2k β
ϕ = ±∫ √ 2α + − dρ + c(t, θ). (2.5.10)
2
r0
ρ ρ
From ansatz (2.5.9) it follows

c(t, θ) = −αt − βθ. (2.5.11)
Therefore we have a two parameter family of solutions

ϕ = ϕ(α, β; θ, r, t) (2.5.12)
of the Hamilton-Jacobi equation. This solution is a complete integral, see an exercise.

According to the theorem of Jacobi set
ϕα = −t0 , ϕβ = −θ0 . (2.5.13)
Then
r
dρ
t − t0 = − ∫ −−−−−−−−−−−. (2.5.14)
2
2
r0 2k β
√ 2α + −
ρ ρ
2
The inverse function r = r(t) , r(0) = r , is the r-coordinate depending on time t , and
0
r
dρ
θ − θ0 = β ∫ −−−−−−−−−−−. (2.5.15)
2
2
r0 2k β
2
ρ √ 2α + −
ρ ρ
2
Substitution τ −1
=ρ yields
1/r
dτ
θ − θ0 = −β ∫ − −−−−−−−−−−−− −
1/r0 √ 2α + 2 k2 τ − β 2 τ 2
2 2
β 1 β 1
−1 −1
k
2 r k
2 r0
= − arcsin ( −−−−−−− ) + arcsin ( −−−−−−− ).
2 2
2αβ 2αβ
√1 + √1 +
4 4
k k
Set
2
β 1
2
−1
k r0
θ1 = θ0 + arcsin ( −−−−−−−) (2.5.16)
2
2αβ
√1 +
4
k
and
−−−−−−−−
2 2
β 2
2αβ
p = , ϵ = √1 + , (2.5.17)
2 4
k k

then
p
−1
r
θ − θ1 = − arcsin( ). (2.5.18)
2
ϵ
It follows
p
r = r(θ) = , (2.5.19)
2
1 −ϵ sin(θ − θ1 )
which is the polar equation of conic sections. It defines an ellipse if 0 ≤ ϵ < 1 , a parabola if ϵ = 1 and a hyperbola if ϵ > 1 ,
see Figure 2.5.2 for the case of an ellipse, where the origin of the coordinate system is one of the focal points of the ellipse.
Figure 2.5.2: The case of an ellipse

For another application of the Jacobi theorem see Courant and Hilbert [4], Vol. 2, pp. 94, where geodesics on an ellipsoid are
studied.
1
Hamilton, William Rowan, 1805--1865
2
Jacobi, Carl Gustav, 1805--1851



2.E: Equations of First Order (Exercises)
calculus sequence.
Q2.1
Suppose u : R 2
↦ R
1
is a solution of
a(x, y)ux + b(x, y)uy = 0. (2.E.1)
Show that for arbitrary H ∈ C also H (u) is a solution.

1
Q2.2
Find a solution u ≢ const. of
ux + uy = 0 (2.E.2)
such that
3 2
graph(u) := {(x, y, z) ∈ R : z = u(x, y), (x, y) ∈ R } (2.E.3)
contains the straight line (0, 0, 1) + s(1, 1, 0), s ∈ R . 1
Q2.3
Let ϕ(x, y) be a solution of
a1 (x, y)ux + a2 (x, y)uy = 0 . (2.E.4)
Prove that level curves SC := {(x, y) : ϕ(x, y) = C = const. } are characteristic curves, provided that ∇ϕ ≠ 0 and
(a , a ) ≠ (0, 0) .
1 2
Q2.4
Prove Proposition 2.2.
Q2.5
Find two different solutions of the initial value problem
ux + uy = 1, (2.E.5)
where the initial data are x 0 (s) =s ,y0 (s) =s ,z0 (s) =s .
Hint: (x 0, y0 ) is a characteristic curve.
Q2.6
Solve the initial value problem
x ux + y uy = u (2.E.6)
with initial data x 0 (s) = s, y0 (s) = 1 ,z

0 (s) , where z is given.
0

Q2.7
2
−x ux + y uy = x u , (2.E.7)
x0 (s) = s, y0 (s) = 1 ,z 0 (s) =e

−s
.
Q2.8
u ux + uy = 1, (2.E.8)
$x_0(s)=s,\ y_0(s)=s$, z 0 (s) = s/2 if 0 < s < 1 .
Q2.9
u ux + u uy = 2, (2.E.9)
x0 (s) = s, y0 (s) = 1 ,z 0 (s) = 1 +s if 0 < s < 1 .
Q2.10
Solve the initial value problem 2 2
ux + uy = 1 + x with given initial data
x0 (s) = 0, y0 (s) = s, u0 (s) = 1, p0 (s) = 1, q0 (s) = 0 , −∞ < s < ∞ .
Q2.11
Find the solution Φ(x, y) of
(x − y)ux + 2y uy = 3x (2.E.10)
such that the surface defined by z = Φ(x, y) contains the curve

$$
C:\ \ x_0(s)=s,\ y_0(s)=1,\ z_0(s)=0,\ s\in{\mathbb R}.
\]
Q2.12
Solve the following initial problem of chemical kinetics.
−k1 x 2
ux + uy = (k0 e + k2 ) (1 − u ) , x > 0, y > 0 (2.E.11)
with the initial data u(x, 0) = 0, u(0, y) = u 0 (y) , where u , 0 < u

0 0 <1 , is given.
Q2.13
Solve the Riemann problem
ux1 + ux2 = 0
u(x1 , 0) = g(x1 )
in Ω = {(x
1 1, x2 ) ∈ R
2
: x1 > x2 } and in Ω2 = {(x1 , x2 ) ∈ R
2
: x1 < x2 } ,
where

ul x1 < 0
g(x1 ) = { (2.E.12)
ur x1 > 0
with constants u l ≠ ur .
Q2.14
Determine the opening angle of the Monge cone, that is, the angle between the axis and the apothem (in German: Mantellinie) of
the cone, for equation
2 2
ux + uy = f (x, y, u), (2.E.13)
where f >0 .
Q2.15
2 2
ux + uy = 1, (2.E.14)
where x 0 (θ) = a cos θ, y0 (θ) = a sin θ, z0 (θ) = 1, p0 (θ) = cos θ ,q
0 (θ) = sin θ if 0 ≤ θ < 2π ,
a = const. > 0 .
Q2.16
Show that the integral ϕ(α, β; θ, r, t), see the Kepler problem, is a complete integral.
Q2.17
−−−−−−
−−−−−
a) Show that S = √−
−
α x + √1 − α y + β , α, β ∈ R
1
, 0 < α < 1 , is a complete integral of S x − √1 − Sy
2
=0 .
b) Find the envelope of this family of solutions.
Q2.18
Determine the length of the half axis of the ellipse
$$
r=\frac{p}{1-\varepsilon^2\sin(\theta-\theta_0)},\ 0\le\varepsilon<1.
\]
Q2.19
Find the Hamilton function H (x, p) of the Hamilton-Jacobi-Bellman differential equation if h = 0 and f = Ax + Bα , where
A, B are constant and real matrices, A : R , B is an orthogonal real n × n -Matrix and p ∈ R is given. The set of
m n n
↦ R
admissible controls is given by

$$
U=\{\alpha\in\mathbb{R}^n:\ \sum_{i=1}^n\alpha_i^2\le1\}\ .
\]
Remark. The Hamilton-Jacobi-Bellman equation is formally the Hamilton-Jacobi equation ut + H (x, ∇u) = 0 , where the
Hamilton function is defined by
H (x, p) := min (f (x, α) ⋅ p + h(x, α)) , (2.E.15)
α∈U
f (x, α) and h(x, α) are given. See for example, Evans [5], Chapter 10.



CHAPTER OVERVIEW
3: CLASSIFICATION
Different types of problems in physics, for example, correspond different types of partial differential equations. The methods how to solve
these equations differ from type to type.
The classification of differential equations follows from one single question: can we calculate formally the solution if sufficiently many
initial data are given? Consider the initial problem for an ordinary differential equation y (x) = f(x, y(x)), y(x ) = y . Then one can
′
0 0
determine formally the solution, provided the function f(x, y) is sufficiently 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 differential equations the related theorem is the Theorem of Cauchy-Kowalevskaya. Even in the case of ordinary differential
equations the situation is more complicated if y is implicitly defined, i. e., the differential equation is F (x, y(x), y (x)) = 0 for a given
′ ′
function F .
Thumbnail: Photograph of the Russian Mathematician Sofja Wassiljewna Kowalewskaja, the first major Russian female mathematician and
responsible for important original contributions to analysis, partial differential equations and mechanics.
CONTRIBUTORS AND ATTRIBUTIONS

3.1: LINEAR EQUATIONS OF SECOND ORDER

3.1.1: NORMAL FORM IN TWO VARIABLES
3.2: QUASILINEAR EQUATIONS OF SECOND ORDER
3.2.1: QUASILINEAR ELLIPTIC EQUATIONS
3.3: SYSTEMS OF FIRST ORDER
3.3.1: EXAMPLES
3.4: SYSTEMS OF SECOND ORDER
3.4.1: EXAMPLES
3.5: THEOREM OF CAUCHY-KOVALEVSKAYA
3.5.1 APPENDIX: REAL ANALYTIC FUNCTIONS
3.E: CLASSIFICATION (EXERCISES)
BACK MATTER
INDEX
1 3/30/2022
CHAPTER OVERVIEW
FRONT MATTER
TITLEPAGE
INFOPAGE
1 3/30/2022
3: Classification
integrated.

3.1: Linear Equations of Second Order
The general nonlinear partial differential equation of second order is
2
F (x, u, Du, D u) = 0, (3.1.1)
where n , u : Ω ⊂ R ↦ R , Du ≡ ∇u and
1
u stands for all second derivatives. The function F is given and sufficiently regular
with respect to its 2n + 1 + n arguments.
2
In this section we consider the case

n
ik
∑ a (x)uxi xk + f (x, u, ∇u) = 0. (3.1.2)
i,k=1
The equation is linear if

n
i
f = ∑ b (x)uxi + c(x)u + d(x). (3.1.3)
i=1
Concerning the classification the main part

n
ik
∑ a (x)uxi xk (3.1.4)
i,k=1
plays the essential role. Suppose u ∈ C , then we can assume, without restriction of generality, that a
2 ik
=a
ki
, since
n n
ik ik ⋆
∑ a uxi xk = ∑ (a ) uxi xk , (3.1.5)
i,k=1 i,k=1
where
ik ⋆
1 ik ki
(a ) = (a +a ). (3.1.6)
2
Consider a hypersurface S in n defined implicitly by χ(x) = 0 , ∇χ ≠ 0 , see Figure 3.1.1
Figure 3.1.1: Initial Manifold S

Assume u and ∇u are given on S .
Problem: Can we calculate all other derivatives of u on S by using differential equation (3.1.2\) and the given data?
We will find an answer if we map S onto a hyperplane S by a mapping
0

λn = χ(x1 , … , xn )
λi = λi (x1 , … , xn ), i = 1, … , n − 1,
for functions λ such that

i
∂(λ1 , … , λn )
det ≠0 (3.1.7)
∂(x1 , … , xn )
in n . It is assumed that i and i are sufficiently regular. Such a mapping λ = λ(x) exists, see an exercise.
The above transform maps S onto a subset of the hyperplane defined by λ n =0 , see Figure 3.1.2.
Figure 3.1.2: Transformed flat manifold S

We will write the differential equation in these new coordinates. Here we use Einstein's convention, that is, we add terms with
repeating indices. Since
u(x) = u(x(λ)) =: v(λ) = v(λ(x)), (3.1.8)
where x = (x 1, … , xn ) and λ = (λ 1, … , λn ) , we get

∂λi
ux = vλ , (3.1.9)
j i
∂xj
2
∂λi ∂λl ∂ λi
uxj xk = vλi λl + vλi .
∂xj ∂xk ∂ xj ∂ xk
Then the differential equation (3.1.2) in the new coordinates is given by

∂λi ∂λl
jk
a (x) vλ λl + terms known on S0 = 0. (3.1.10)
i
∂xj ∂xk
Since v (λ , … , λ
λk 1 n−1 , , , are known, see (3.1.9), it follows that v
0) n λk λl , l = 1, … , n − 1 , are known on S . Thus we know all
0
second derivatives v λi λj on S with the only exception of v

0 . λn λn
We recall that, provided v is sufficiently regular,

vλ λl (λ1 , … , λn−1 , 0) (3.1.11)
k
is the limit of
vλ (λ1 , … , λl + h, λl+1 , … , λn−1 , 0) − vλ (λ1 , … , λl , λl+1 , … , λn−1 , 0)
k k
(3.1.12)
h
as h → 0 .
Then the differential equation can be written as

n
jk
∂λn ∂λn
∑ a (x) vλn λn = terms known on S0 . (3.1.13)
∂xj ∂xk
j,k=1
It follows that we can calculate vλn λn if

n
ij
∑ a (x)χx χx ≠0 (3.1.14)
i j
i,j=1
on S . This is a condition for the given equation and for the given surface S .
Definition. The differential equation
n
ij
∑ a (x)χxi χxj = 0 (3.1.15)
i,j=1
is called it characteristic differential equation associated to the given differential equation (3.1.2).
If χ, ∇χ ≠ 0 , is a solution of the characteristic differential equation, then the surface defined by χ =0 is called characteristic
surface.
Remark. The condition (3.1.14) is satisfied for each χ with ∇χ ≠ 0 if the quadratic matrix (a (x)) is positive or negative ij
definite for each x ∈ Ω, which is equivalent to the property that all eigenvalues are different from zero and have the same sign.
This follows since there is a λ(x) > 0 such that, in the case that the matrix (a ) is poitive definite, ij
ij 2
∑ a (x)ζi ζj ≥ λ(x)|ζ | (3.1.16)
i,j=1
for all ζ ∈ R . Here and in the following we assume that the matrix (a ij
) is real and symmetric.
The characterization of differential equation (3.1.2) follows from the signs of the eigenvalues of (a ij
(x)) .
Definition. The differential equation (3.1.2) is said to be of type (α, β, γ) at x ∈ Ω if α eigenvalues of ij
(a )(x) are positive, β
eigenvalues are negative and γ eigenvalues are zero (α + β + γ = n ).

In particular, the equation is called
elliptic if it is of type (n, 0, 0) or of type (0, n, 0), that is, all eigenvalues are different from zero and have the same sign,\\
parabolic if it is of type (n − 1, 0, 1) or of type (0, n − 1, 1) , that is, one eigenvalue is zero and all the others are different from
zero and have the same sign,
hyperbolic if it is of type (n − 1, 1, 0) or of type (1, n − 1, 0) , that is, all eigenvalues are different from zero and one eigenvalue
has another sign than all the others.
Remarks:
1. According to this definition there are other types aside from elliptic, parabolic or hyperbolic equations.
2. The classification depends in general on x ∈ Ω. An example is the Tricomi equation, which appears in the theory of transsonic
flows,
y uxx + uyy = 0. (3.1.17)
This equation is elliptic if y > 0 , parabolic if y = 0 and hyperbolic for y < 0 .
Examples:

Example 3.1.1:
The Laplace equation in R is △u = 0 , where

3
△u := uxx + uyy + uzz . (3.1.18)
This equation is elliptic since for every manifold S given by {(x, y, z) : χ(x, y, z) = 0} , where χ is an arbitrary sufficiently
regular function such that ∇χ ≠ 0 , all derivatives of u are known on S , provided u and ∇u are known on S .
Example 3.1.2:
The wave equation u = u + u tt xx yy + uzz , where u = u(x, y, z, t) , is hyperbolic. Such a type describes oscillations of
mechanical structures, for example.
Example 3.1.3:
The heat equation u = u + u + u , where u = u(x, y, z, t), is

t xx yy zz
parabolic. It describes, for example, the propagation of heat in a domain.
Example 3.1.4:
Consider the case that the (real) coefficients a in equation (3.1.2) are {\it constant}. We recall that the matrix A = (a ) is
ij ij
symmetric, that is, A = A . In this case, the transform to principle axis leads to a normal form from which the classification of
T
the equation is obviously. Let U be the associated orthogonal matrix, then

λ1 0 ⋯ 0
⎛ ⎞
T ⎜0 λ2 ⋯ 0 ⎟
U AU = ⎜ ⎟. (3.1.19)
⎜... ... ... ...⎟
⎝ ⎠
0 0 ⋯ λn
Here is U = (z1 , … , zn ) , where z , l = 1, … , n, is an orthonormal system of eigenvectors to the eigenvalues λ .

l l
Set y = U T
x and v(y) = u(U y) , then
n n
ij
∑ a uxi xj = ∑ λi vy yj . (3.1.20)
i
i,j=1 i=1



3.1.1: Normal Form in Two Variables
a(x, y)uxx + 2b(x, y)uxy + c(x, y)uyy + terms of lower order = 0 (3.1.1.1)
in Ω ⊂ R . The associated characteristic differential equation is

2
2 2
aχx + 2b χx χy + c χy = 0. (3.1.1.2)
We show that an appropriate coordinate transform will simplify equation (3.1.1.1) sometimes in such a way that we can solve the
transformed equation explicitly.
Let z = ϕ(x, y) be a solution of (3.1.1.2). Consider the level sets {(x, y) : ϕ(x, y) = const. } and assume ϕ ≠ 0 at a point y
(x0 , y0 ) of the level set. Then there is a function y(x) defined in a neighborhood of x such that ϕ(x, y(x)) = const. It follows
0
$$y'(x)=-\dfrac{\phi_x}{\phi_y},\]
which implies, see the characteristic equation (3.1.1.2),
′2 ′
ay − 2b y + c = 0. (3.1.1.3)
Then, provided a ≠ 0 , we can calculate μ := y from the (known) coefficients a , b and c :

′
1 − − −−−−
2
μ1,2 = (b ± √ b − ac ) . (3.1.1.4)
a
These solutions are real if and only of ac − b 2

≤0 .
Equation (3.1.1.1) is hyperbolic if ac − b 2
<0 , parabolic if ac − b
2
=0 and ellipticif 2
ac − b >0 . This follows from an easy
discussion of the eigenvalues of the matrix
$$\left(
a b
(3.1.1.1)
b c
\right),\]
see an exercise.
Normal form of a hyperbolic equation

Let ϕ and ψ are solutions of the characteristic equation (3.1.1.2) such that
ϕx
′
y ≡ μ1 = −
1
ϕy
ψx
′
y ≡ μ2 = − ,
2
ψy
where μ and μ are given by (3.1.1.4). Thus ϕ and ψ are solutions of the linear homogeneous equations of first order
1 2
ϕx + μ1 (x, y)ϕy = 0 (3.1.1.5)
ψx + μ2 (x, y)ψy = 0. (3.1.1.6)
Assume ϕ(x, y), ψ(x, y) are solutions such that ∇ϕ ≠ 0 and ∇ψ ≠ 0 , see an exercise for the existence of such solutions.
Consider two families of level sets defined by ϕ(x, y) = α and ψ(x, y) = β, see Figure 3.1.1.1.
alt
Figure 3.1.1.1: Level sets

These level sets are characteristic curves of the partial differential equations (3.1.1.5) and (3.1.1.6), respectively, see an exercise of
the previous chapter.
Lemma. (i) Curves from different families can not touch each other

(ii) ϕ x ψy − ϕy ψx ≠ 0 .
Proof. (i):
2 − − −−−−
′ ′ 2
y −y ≡ μ2 − μ1 = − √ b − ac ≠ 0. (3.1.1.2)
2 1
a
(ii):
$$
\mu_2-\mu_1=\dfrac{\phi_x}{\phi_y}-\dfrac{\psi_x}{\psi_y}.
\]
□
Proposition 3.1. The mapping ξ = ϕ(x, y), η = ψ(x, y) transforms equation (3.1.1.1) into
vξη = lower order terms, (3.1.1.7)
where v(ξ, η) = u(x(ξ, η), y(ξ, η)).}

Proof. The proof follows from a straightforward calculation.
ux = vξ ϕx + vη ψx
uy = vξ ϕy + vη ψy
2 2
uxx = vξξ ϕx + 2 vξη ϕx ψx + vηη ψx + lower order terms
uxy = vξξ ϕx ϕy + vξη (ϕx ψy + ϕy ψx ) + vηη ψx ψy + lower order terms
2 2
uyy = vξξ ϕy + 2 vξη ϕy ψy + vηη ψy + lower order terms.
Thus
auxx + 2b uxy + c uyy = α vξξ + 2β vξη + γ vηη + l. o. t. , (3.1.1.3)
where
2 2
α : = aϕx + 2b ϕx ϕy + c ϕy
β : = aϕx ψx + b(ϕx ψy + ϕy ψx ) + c ϕy ψy
2 2
γ : = aψx + 2b ψx ψy + c ψy .
The coefficients α and γ are zero since ϕ and ψ are solutions of the characteristic equation. Since
2 2 2
αγ − β = (ac − b )(ϕx ψy − ϕy ψx ) , (3.1.1.4)
it follows from the above lemma that the coefficient β is different from zero.
□
Example 3.1.1.1:
uxx − uyy = 0. (3.1.1.5)
The associated characteristic differential equation is

2 2
χx − χy = 0. (3.1.1.6)

Since μ1 = −1 and μ
2 =1 , the functions ϕ and ψ satisfy differential equations
ϕx + ϕy = 0
ψx − ψy = 0.
Solutions with ∇ϕ ≠ 0 and ∇ψ ≠ 0 are

ϕ = x − y, ψ = x + y. (3.1.1.7)
Then the mapping
ξ = x − y, η = x + y (3.1.1.8)
leads to the simple equation
vξη (ξ, η) = 0. (3.1.1.9)
Assume v ∈ C is a solution, then v

2
ξ = f1 (ξ) for an arbitrary C function f
1
1 (ξ) . It follows
ξ
v(ξ, η) = ∫ f1 (α) dα + g(η), (3.1.1.10)

0
where g is an arbitrary C function. Thus each C -solution of the differential equation can be written as
2 2
(⋆) v(ξ, η) = f (ξ) + g(η) ,

where f , g ∈ C . On the other hand, for arbitrary C -functions f , g the function (⋆) is a solution of the differential equation
2 2
= 0 . Consequently every C -solution of the original equation u = 0 is given by

2
vξη −u xx yy
u(x, y) = f (x − y) + g(x + y), (3.1.1.11)
where f , g ∈ C . 2



3.2: Quasilinear Equations of Second Order
Here we consider the equation
n
ij
∑ a (x, u, ∇u)uxi xj + b(x, u, ∇u) = 0 (3.2.1)
i,j=1
in a domain Ω ⊂ R , where u : Ω ↦ R . We assume that a

1 ij
=a
ji
.
As in the previous section we can derive the characteristic equation
n
ij
∑ a (x, u, ∇u)χxi χxj = 0. (3.2.2)
i,j=1
In contrast to linear equations, solutions of the characteristic equation depend on the solution considered.



3.2.1: Quasilinear Elliptic Equations
There is a large class of quasilinear equations such that the associated characteristic equation has no solution χ, ∇χ ≠ 0 .
Set
$$
U=\{(x,z,p):\ x\in\Omega,\ z\in\mathbb{R}^1,\ p\in\mathbb{R}\}.
\]
Definition. The quasilinear equation (3.2.1) is called elliptic if the matrix (a ij
(x, z, p)) is positive definite for each (x, z, p) ∈ U .
Assume equation (3.2.1) is elliptic and let λ(x, z, p) be the minimum and Λ(x, z, p) the maximum of the eigenvalues of (a ij
) , then
n
2 ij 2
0 < λ(x, z, p)|ζ | ≤ ∑ a (x, z, p)ζi ζj ≤ Λ(x, z, p)|ζ | (3.2.1.1)
i,j=1
for all ζ ∈ R .
Definition. Equation (3.2.1) is called uniformly elliptic if Λ/λ is uniformly bounded in U .
An important class of elliptic equations which are not uniformly elliptic (non-uniformly elliptic) is
n ⎛ ⎞
∂ uxi
∑ ⎜ −−−−−−−− ⎟ + lower order terms = 0. (3.2.1.1)
∂xi 2
i=1 ⎝ √ 1 + |∇u| ⎠
The main part is the minimal surface operator (left hand side of the minimal surface equation). The coefficients a are ij
−1/2 pi pj
ij 2
a (x, z, p) = (1 + |p | ) ( δij − ), (3.2.1.2)
2
1 + |p|
δij denotes the Kronecker delta symbol. It follows that

1 1
λ = , Λ = . (3.2.1.3)
3/2 1/2
2 2
(1 + |p | ) (1 + |p | )
Thus equation (3.2.1.1) is not uniformly elliptic.

The behavior of solutions of uniformly elliptic equations is similar to linear elliptic equations in contrast to the behavior of
solutions of non-uniformly elliptic equations.
Typical examples for non-uniformly elliptic equations are the minimal surface equation and the capillary equation.



3.3: Systems of First Order
Consider the quasilinear system
n
k
∑ A (x, u)uuk + b(x, u) = 0, (3.3.1)
k=1
where A are m × m -matrices, sufficiently regular with respect to their arguments, and
k
$$
u=\left(
u1
(3.3.2)
⋮
um
\right),\ \
u_{x_k}=\left(
u1,x
k
(3.3.3)
⋮
um,x
k
\right),\ \
b=\left(
b1
(3.3.4)
⋮
bm
\right).
\]
We ask the same question as above: can we calculate all derivatives of u in a neighborhood of a given hypersurface S in R defined
by χ(x) = 0 , ∇χ ≠ 0 , provided u(x) is given on S ?
For an answer we map S onto a flat surface S by using the mapping
0 λ = λ(x) of Section 3.1 and write equation (3.3.1) in new
coordinates. Set v(λ) = u(x(λ)) , then
$$\sum_{k=1}^nA^k(x,u)\chi_{x_k}v_{\lambda_n}=\mbox{terms known on}\ \mathcal{S}_0.\]
We can solve this system with respect to v , provided that
λn
$$\det\left(\sum_{k=1}^nA^k(x,u)\chi_{x_k}\right)\not=0\]
on S .
Definition. Equation
$$\det\left(\sum_{k=1}^nA^k(x,u)\chi_{x_k}\right)=0\]
is called characteristic equation associated to equation (3.3.1) and a surface S : χ(x) = 0 , defined by a solution χ, ∇χ ≠ 0 , of this
characteristic equation is said to be characteristic surface.
Set
$$C(x,u,\zeta)=\det\left(\sum_{k=1}^nA^k(x,u)\zeta_k\right)\]
for ζ k ∈ R .
Definition.

i. The system (3.3.1) is hyperbolic at (x, u(x)) if there is a regular linear mapping ζ = Qη , where η = (η 1, … , ηn−1 , κ) , such
that there exists m {\it real} roots κ = κ (x, u(x), η , … , η ), k = 1, … , m, of
k k 1 n−1
D(x, u(x), η1 , … , ηn−1 , κ) = 0 (3.3.5)
for all (η
1, , where
… , ηn−1 )
D(x, u(x), η1 , … , ηn−1 , κ) = C (x, u(x), x, Qη). (3.3.6)
ii. System (3.3.1) is parabolic if there exists a regular linear mapping ζ = Qη such that D is independent of κ , that is, D depends
on less than n parameters.
iii. System (3.3.1) is elliptic if C (x, u, ζ) = 0 only if ζ = 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
Example 3.3.1.1: Beltrami Equations
W ux − b vx − c vy = 0 (3.3.1.1)
W uy + avx + b vy = 0, (3.3.1.2)
where W , a, b, c are given functions depending of (x, y), W ≠0 and the matrix
$$
\left(
a b
(3.3.1.1)
b c
\right)
\]
is positive definite.
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 quasiconform mapping, see for example [9], Chapter 12, for an application to partial differential
equations.
Set
$$
A^1=\left(
W −b
(3.3.1.2)
0 a
\right),\ \
A^2=\left(
0 −c
(3.3.1.3)
W b
\right).
\]
Then the system (3.3.1.1), (3.3.1.2) can be written as
$$
A^1\left(
ux
(3.3.1.4)
vx
\right)+
A^2\left(
uy
(3.3.1.5)
vy
\right)=\left(
0
(3.3.1.6)
0
\right).
\]
Thus,

∣ W ζ1 −b ζ1 − c ζ2 ∣
2 2
C (x, y, ζ) = ∣ ∣ = W (aζ + 2b ζ1 ζ2 + c ζ ),
1 2
∣ W ζ2 aζ1 + b ζ2 ∣
which is different from zero if ζ ≠ 0 according to the above assumptions. Thus the Beltrami system is elliptic.
Example 3.3.1.2: Maxwell Equations

The Maxwell equations in the isotropic case are
c rotx H = λE + ϵEt (3.3.1.3)
c rotx E = −μHt , (3.3.1.4)
where
E = (e1 , e2 , e3 )
T
electric field strength, e = e (x, t) , x = (x
i i 1, x2 , x3 ) ,
H = (h1 , h2 , h3 )
T
magnetic field strength, h = h (x, t) ,
i i
c speed of light,
λ specific conductivity,
ϵ dielectricity constant,
μ magnetic permeability.
Here c, λ, ϵ and μ are positive constants.

Set p 0 = χt , pi = χxi , i = 1, … 3, then the characteristic differential equation is
$$
\left|
ϵp0 /c 0 0 0 p3 −p2
0 ϵp0 /c 0 −p3 0 p1
0 0 ϵp0 /c p2 −p1 0
(3.3.1.7)
0 −p3 p2 μp0 /c 0 0
p3 0 −p1 0 μp0 /c 0
−p2 p1 0 0 0 μp0 /c
\right|=0.
\]
The following manipulations simplifies this equation:
i. multiply the first three columns with μp /c, 0
ii. multiply the 5th column with −p and the the 6th column with p and add the sum to the 1st column,
3 2
iii. multiply the 4th column with p and the 6th column with −p and add the sum to the 2th column,
3 1
iv. multiply the 4th column with −p and the 5th column with p and add the sum to the 3th column,
2 1
v. expand the resulting determinant with respect to the elements of the 6th, 5th and 4th row.
We obtain
$$
\left|
2
q +p p1 p2 p1 p3
1
2
p1 p2 q +p p2 p3 (3.3.1.8)
2
2
p1 p3 p2 p3 q +p
3
\right|=0,
\]
where
$$
q:=\frac{\epsilon\mu}{c^2}p_0^2-g^2

\]
with g 2
:= p
2
1
+p
2
2
2
+p
3
. The evaluation of the above equation leads to q 2 2
(q + g ) = 0 , i. e.,
$$
\chi_t^2\left(\frac{\epsilon\mu}{c^2}\chi_t^2-|\nabla_x\chi|^2\right)=0.
\]
It follows immediately that Maxwell equations are a hyperbolic system, see an exercise.
There are two solutions of this characteristic equation. The first one are characteristic surfaces S(t) , defined by χ(x, t) = 0 ,
which satisfy χ = 0 . These surfaces are called stationary waves The second type of characteristic surfaces are defined by
t
solutions of
$$
\frac{\epsilon\mu}{c^2}\chi_t^2=|\nabla_x\chi|^2.
\]
Functions defined by χ = f (n ⋅ x − V 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/ √ϵμ .
The associated characteristic surfaces S(t) are defined by
$$
\chi(x,t)\equiv f(n\cdot x-Vt)=0,
\]
here we assume that 0 is in he range of f : R ↦ R . Thus, S(t) is defined by n ⋅ x − V t = c , where c is a fixed constant. It
1 1
follows that the planes S(t) with normal n move with speed V in direction of n , see Figure 3.3.1.1.
Figure 3.3.1.1: d ′
(t) is the speed of plane waves
V is called speed of the plane wave S(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 fields free of sources. In fact, it is sufficient to assume that this assumption is satisfied at a fixed
time t only, see an exercise.

0
Since
$$
\text{rot}_x\ \text{rot}_x\ A=\mbox{grad}_x\ \text{div}_x\ A-\triangle_xA
\]
for each C -vector field A , it follows from Maxwell equations the uncoupled system
2
ϵμ λμ
△x E = Ett + Et
2
c c2
ϵμ λμ
△x H = Htt + Ht .
2 2
c c
Example 3.3.1.3: Equations of Gas Dynamics

Consider the following quasilinear equations of first order.

$$
v_t+(v\cdot\nabla_x)\ v+\frac{1}{\rho} \nabla_x p =f\ \ \ \mbox{(Euler equations)}.
\]
Here is
the vector of speed, v = v (x, t) , x = (x , x
v = (v1 , v2 , v3 ) i i 1 2, x3 ) ,
p pressure, p = (x, t) ,
ρ density, ρ = ρ(x, t) ,
f = (f , f , f ) density of the external force, f = f (x, t) ,

1 2 3 i i
(v ⋅ ∇x )v ≡ (v ⋅ ∇x v1 , v ⋅ ∇x v2 , v ⋅ ∇x v3 ))
T
.
The second equation is
$$
\rho_t+v\cdot\nabla_x\rho+\rho\ \text{div}_x\ v=0\ \ \ \mbox{(conservation of mass)}.
\]
Assume the gas is compressible and that there is a function (state equation)
$$
p=p(\rho),
\]
where p (ρ) > 0 if ρ > 0 . Then the above system of four equations is
′
1 ′
vt + (v ⋅ ∇)v + p (ρ)∇ρ = f (3.3.1.5)
ρ
ρt + ρ div v + v ⋅ ∇ρ = 0, (3.3.1.6)
where ∇ ≡ ∇ and div ≡ div , i. e., these operators apply on the spatial variables only.
x x
The characteristic differential equation is here

$$
\left|
dχ 1 ′
0 0 p χx1
dt ρ
dχ 1 ′
0 0 p χx2
dt ρ
(3.3.1.9)
dχ 1 ′
0 0 p χx3
dt ρ
dχ
ρχx1 ρχx2 ρχx3
dt
\right|=0,
\]
where
$$\dfrac{d\chi}{dt}:=\chi_t+(\nabla_x\chi)\cdot v. \]
Evaluating the determinant, we get the characteristic differential equation
2 2
dχ dχ 2
′
( ) (( ) − p (ρ)| ∇x χ| ) = 0. (3.3.1.7)
dt dt
This equation implies consequences for the speed of the characteristic surfaces as the following consideration shows.
Consider a family S(t) of surfaces in R defined by χ(x, t) = c , where
3
x ∈ R and c is a fixed constant. As usually, we assume that ∇ χ ≠ 0 .

3
x
One of the two normals on S(t) at a point of the surface S(t) is given by, see an exercise,
∇x χ
n = . (3.3.1.8)
| ∇x χ|

Let Q 0 ∈ S(t0 ) and let Q ∈ S(t ) be a point on the line defined by Q
1 1 0 + sn , where n is the normal (3.3.1.8) on S(t 0) at Q
0
and t0 < t1 ,t
1 −t 0 small, see Figure 3.3.1.2.
3.3.1.2: Definition of the speed of a surface

Definition. The limit
| Q1 − Q0 |
P = lim (3.3.1.10)
t1 →t0 t1 − t0
is called speed of the surface S(t) .

Proposition 3.2. The speed of the surface S(t) is
χt
P =− . (3.3.1.11)
| ∇x χ|
Proof. The proof follows from χ(Q 0, t0 ) = 0 and χ(Q 0 + dn, t0 + △t) = 0 , where d = |Q 1 − Q0 | and △t = t1 − t0 .
□
Set v := v ⋅ n which is the component of the velocity vector in direction n .

n
From ({3.3.1.8) we get

$$
v_n=\frac{1}{|\nabla_x\chi|}v\cdot \nabla_x\chi.
\]
Definition. V := P − vn , the difference of the speed of the surface and the speed of liquid particles, is called relative speed.
Figure 3.3.1.3: Definition of relative speed

Using the above formulas for P and v it follows n
χt v ⋅ ∇x χ 1 dχ
V = P − vn = − − =− . (3.3.1.12)
| ∇x χ| | ∇x χ| | ∇x χ| dt
Then, we obtain from the characteristic equation (3.3.1.7) that

2 2 2 2 ′ 2
V | ∇x χ| (V | ∇x χ| − p (ρ)| ∇x χ| ) = 0. (3.3.1.13)
An interesting conclusion is that there are two relative speeds: V =0 or V 2 ′

= p (ρ) .
−− −−
Definition. ′
√p (ρ) is called speed of sound.



3.4: Systems of Second Order
Here we consider the system
n
kl
∑ A (x, u, ∇u)uxk xl + lower order terms = 0, (3.4.1)
k,l=1
where A are (m × m) matrices and u = (u , … , u ) . We assume A = A , which is no restriction of generality provided

kl
1 m
T kl lk
$u\in C^2$ is satisfied.

As in the previous sections, the classification follows from the question whether or not we can calculate formally the solution from
the differential equations, if sufficiently many data are given on an initial manifold. Let the initial manifold S be given by
χ(x) = 0 and assume that ∇χ ≠ 0 . The mapping x = x(λ), see previous sections, leads to
kl
∑ A χxk χxl vλn λn = terms known on S, (3.4.2)
k,l=1
where v(λ) = u(x(λ)) .

The characteristic equation is here
n
kl
det ( ∑ A χxk χxl ) = 0. (3.4.3)
k,l=1
If there is a solution χ with ∇χ ≠ 0 , then it is possible that second derivatives are not continuous in a neighborhood of S .
Definition. The system is called elliptic if
n
kl
det ( ∑ A ζk ζl ) ≠ 0 (3.4.4)
k,l=1
for all ζ ∈ R , ζ ≠ 0 .



3.4.1: Examples
Example 3.4.1.1: Navier-Stokes equations
The Navier-Stokes system for a viscous incompressible liquid is
1
vt + (v ⋅ ∇x )v = − ∇x p + γ △x v
ρ
divx v = 0,
where
ρ is the (constant and positive) density of liquid,
γ is the (constant and positive) viscosity of liquid,
v = v(x, t) velocity vector of liquid particles, x ∈ R or in R ,

3 2
p = p(x, t) pressure.
The problem is to find solutions v, p of the above system.
Example 3.4.2.1: Linear elasticity

Consider the system
2
∂ u
ρ = μ△x u + (λ + μ)∇x (divx u) + f . (3.4.1.1)
2
∂t
Here is, in the case of an elastic body in R ,3
u(x, t) = (u (x, t), u (x, t), u (x, t)) displacement vector,

1 2 3
f (x, t) density of external force,
ρ (constant) density,
λ, μ (positive) Lamé constants.
The characteristic equation is det C =0 where the entries of the matrix C are given by
2 2
cij = (λ + μ)χxi χxj + δij (μ| ∇x χ| − ρχ ) . (3.4.1.2)
t
The characteristic equation is

2
2 2 2 2
((λ + 2μ)| ∇x χ| − ρχ ) (μ| ∇x χ| − ρχ ) = 0. (3.4.1.3)
t t
It follows that two different speeds P of characteristic surfaces S(t) , defined by

χ(x, t) = const. , are possible, namely
−−−−−−
−
−
λ + 2μ μ
P1 = √ , and P2 = √ . (3.4.1.4)
ρ ρ
We recall that P = −χt /| ∇x χ| .



3.5: Theorem of Cauchy-Kovalevskaya
Consider the quasilinear system of first order (3.3.1) of Section 3.3. Assume an initial manifolds S is given by χ(x) = 0 , ∇χ ≠ 0 ,
and suppose that χ is not characteristic. Then, see Section 3.3, the system (3.3.1) can be written as
n−1
i
ux = ∑ a (x, u)ux + b(x, u) (3.5.1)
n i
i=1
u(x1 , … , xn−1 , 0) = f (x1 , … , xn−1 ) (3.5.2)
Here is u = (u 1, … , um )
T
, b = (b
1, … , bn )
T
and a are (m × m) -matrices.
i
We assume that a , b and f are in C with respect to their arguments. From (3.5.1) and (3.5.2) it follows that we can calculate
i ∞
formally all derivatives D u in a neighborhood of the plane {x : x = 0} , in particular in a neighborhood of 0 ∈ R . Thus we

α
n
have a formal power series of u(x) at x = 0 :

$$u(x)\sim \sum\frac{1}{\alpha !}D^\alpha u(0) x^\alpha.\]
For notations and definitions 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 neighborhood of 0 ∈ R ?
ii. Is a convergent power series a solution of the initial value problem (3.5.1), (3.5.2)?
Remark. Quite different 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 finite sum ∑ ϕ (x) of functions ϕ . In general, the infinite sum
N
i=0 i i
ϕ (x) does not converge, in contrast to the power series method of this section. See [15] for some asymptotic formulas in
∞
∑ i
i=0
capillarity.
Theorem 3.1. (Cauchy-Kovalevskaya). There is a neighborhood of 0 ∈ R such there is a real analytic solution of the initial value
problem (3.5.1), (3.5.2). This solution is unique in the class of real analytic functions.
Proof. The proof is taken from F. John \cite{John}. We introduce u − f 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 . Then
⋆ ⋆
n
$$u^\star_{x_n}=1,\ u^\star_{x_k}=0,\ k=1,\ldots,n-1,\ u^\star(x_1,\ldots,x_{n-1},0)=0\]

and the extended system (3.5.1), (3.5.2) is
$$
\left(
u1,xn
⋮
(3.5.3)
um,xn
⋆
ux
n
\right)=
\sum_{i=1}^{n-1}\left(
i
a 0
(3.5.4)
0 0
\right)
\left(
u1,x
i
⋮
(3.5.5)
um,xi
⋆
ux
i

\right)+
\left(
b1
⋮
(3.5.6)
bm
\right),
\]
where the associated initial condition is u(x 1, … , xn−1 , 0) = 0 .
The new u is u = (u1 , … , um )
T
, the new a
i
are i
a (x1 , … , xn−1 , u1 , … , um , u )
⋆
and the new b is
b = (x1 , … , xn−1 , u1 , … , um , u )
⋆ T
.
Thus we are led to an initial value problem of the type
n−1 N
i
uj,xn = ∑ ∑ a (z)uk,xi + bj (z), j = 1, … , N (3.5.7)
jk
i=1 k=1
uj (x) = 0 if xn = 0, (3.5.8)
where j = 1, … , N and z = (x 1, … , xn−1 , u1 , … , uN ) .

The point here is that a i
jk
and b are independent of x . This fact simplifies the proof of the theorem.
j n
From (3.5.7) and (3.5.8) we can calculate formally all D β

uj . Then we have formal power series for u : j
$$u_j(x)\sim \sum_\alpha c_\alpha^{(j)}x^\alpha,\]

where
$$c_\alpha^{(j)}=\frac{1}{\alpha!}D^\alpha u_j(0).\]
We will show that these power series are (absolutely) convergent in a neighborhood of 0 ∈ R , i.e., they are real analytic functions,
see the appendix for the definition of real analytic functions. Inserting these functions into the left and into the right hand side of (
3.5.7) 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 coefficients caused by the calculation of the derivatives D u (0) from (3.5.7). It follows that u (x), α
j j
j = 1, … , n , defined by its formal power series are solutions of the initial value problem (3.5.7), (3.5.8).
Set
$$d=\left(\frac{\partial}{\partial z_1},\ldots,\frac{\partial}{\partial z_{N+n-1}}\right)\]
Lemma A. Assume u ∈ C ∞
in a neighborhood of 0 ∈ R . Then
α β i γ
D uj (0) = Pα (d a (0), d bj (0)) , (3.5.9)
jk
where |β|, |γ| ≤ |α| and P are polynomials in the indicated arguments with non-negative integers as coefficients which are
α
independent of a and of b .
i
Proof. It follows from equation (3.5.7) that

α β i γ δ
Dn D uj (0) = Pα (d a (0), d bj (0), D uk (0)). (3.5.10)
jk
Here is D n = ∂/∂ xn and α, β, γ, δ satisfy the inequalities

$$|\beta|,\ |\gamma|\le|\alpha|,\ \ |\delta|\le|\alpha|+1,\]
and, which is essential in the proof, the last coordinates in the multi-indices α = (α , … , α ) , δ = (δ , … , δ ) satisfy δ 1 n 1 n n ≤ αn
since the right hand side of (3.5.7) is independent of x . n
Moreover, it follows from (3.5.7) that the polynomials P have integers as coefficients. The initial condition (3.5.8) implies
α

α
D uj (0) = 0, (3.5.11)
where , that is, α = 0 . Then, the proof is by induction with respect to α . The induction starts with
α = (α1 , … , αn−1 , 0) n n
αn = 0 , then we replace $D^\delta u_k(0)$ in the right hand side of (3.5.10) by (3.5.11), that is by zero. Then it follows from (
3.5.10) that
$$D^\alpha u_j(0)=P_\alpha (d^\beta a_{jk}î(0),d^\gamma b_j(0),D^\delta u_k(0)),\]

where α = (α 1, … , αn−1 , 1) .
□
Definition. Let f = (f1 , … , fm ) ,F = (F1 , … , Fm ) ,f i = fi (x) ,F i = Fi (x) , and f , F ∈ C

∞
. We say f is majorized by F if
α α
|D fk (0)| ≤ D Fk (0), k = 1, … , m (3.5.12)
for all α . We write f << F , if f is majorized by F .

Definition. The initial value problem
n−1 N
i
Uj,xn = ∑ ∑ A (z)Uk,xi + Bj (z) (3.5.13)
jk
i=1 k=1
Uj (x) = 0 if xn = 0, (3.5.14)
j = 1, … , N ,A
, B real analytic, is called majorizing problem to (3.5.7), (3.5.8) if
i
jk j
$$
a_{jk}î<<A_{jk}î\ \ \mbox{and}\ b_j<<B_j.
\]
Lemma B. The formal power series
1 α α
∑ D uj (0)x , (3.5.15)
α!
α
where D u (0) are defined in Lemma A, is convergent in a neighborhood of 0 ∈ R if there exists a majorizing problem which has
α
j
a real analytic solution U in x = 0 , and

α α
|D uj (0)| ≤ D Uj (0). (3.5.16)
Proof. It follows from Lemma A and from the assumption of Lemma B that
α β i γ
|D uj (0)| ≤ Pα (| d a (0)|, | d bj (0)|)
jk
β i γ α
≤ Pα (| d A (0)|, | d Bj (0)|) ≡ D Uj (0).
jk
The formal power series

1 α α
∑ D uj (0)x , (3.5.17)
α!
α
is convergent since
1 α α
1 α α
∑ |D uj (0)x | ≤∑ D Uj (0)| x |. (3.5.18)
α! α!
α α
The right hand side is convergent in a neighborhood of x ∈ R by assumption.

□
Lemma C. There is a majorising problem which has a real analytic solution.

Proof. Since a (z) , b (z) are real analytic in a neighborhood of z = 0 it follows from Proposition A5 of the appendix to this
i
ij j
section that there are positive constants M and r such that all these functions are majorized by
Mr
. (3.5.19)
r − z1 − … − zN +n−1
Thus a majorizing problem is

n−1 N
Mr
Uj,x = (1 + ∑ ∑ Uk,x )
n i
r − x1 − … − xn−1 − U1 − … − UN
i=1 k=1
Uj (x) = 0 if xn = 0,
j = 1, … , N .
The solution of this problem is
Uj (x1 , … , xn−1 , xn ) = V (x1 + … + xn−1 , xn ), j = 1, … , N , (3.5.20)
where V (s, t), s = x 1 + … + xn−1 , t = x , is the solution of the Cauchy initial value problem
n
Mr
Vt = (1 + N (n − 1)Vs ) ,
r −s−NV
V (s, 0) = 0.
which has the solution, see an exercise,

−−−−−−−−−−−−−−−
1 2
V (s, t) = (r − s − √ (r − s) − 2nM N rt ) . (3.5.21)
Nn
This function is real analytic in (s, t) at (0, 0). It follows that Uj (x) are also real analytic functions. Thus the Cauchy-
Kovalevskaya theorem is shown.
□
Example 3.5.1: Ordinary differential equations

Consider the initial value problem
′
y (x) = f (x, y(x))
y(x0 ) = y0 ,
where x ∈ R and y ∈ R are given. Assume f (x, y) is real analytic in a neighborhood of (x , y ) ∈ R × R . Then it
0
1
0 0 0
1
follows from the above theorem that there exists an analytic solution y(x) of the initial value problem in a neighborhood of x . 0
This solution is unique in the class of analytic functions according to the theorem of Cauchy-Kovalevskaya. From the Picard-
Lindel\"of theorem it follows that this analytic solution is unique even in the class of C -functions. 1
Example 3.5.2: Partial differential equations of second order

Consider the boundary value problem for two variables

uyy = f (x, y, u, ux , uy , uxx , uxy )
u(x, 0) = ϕ(x)
uy (x, 0) = ψ(x).
We assume that ϕ, ψ are analytic in a neighborhood of x = 0 and that f is real analytic in a neighbourhood of
′ ′
(0, 0, ϕ(0), ϕ (0), ψ(0), ψ (0)). (3.5.22)
There exists a real analytic solution in a neighborhood of 0 ∈ R of the above initial value problem.
2
In particular, there is a real analytic solution in a neighborhood of 0 ∈ R of the initial value problem 2
△u = 1
u(x, 0) = 0
uy (x, 0) = 0.
The proof follows by writing the above problem as a system. Set p = u , q = u , r = u x y xx ,

s =u , t = u , then
xy yy
t = f (x, y, u, p, q, r, s). (3.5.23)
Set U = (u, p, q, r, s, t)
T
, b = (q, 0, t, 0, 0, f
y + fu q + fq t)
T
and
0 0 0 0 0 0
⎛ ⎞
⎜0 0 1 0 0 0 ⎟
⎜ ⎟
⎜0 0 0 0 0 0 ⎟
A =⎜ ⎟. (3.5.24)
⎜ ⎟
⎜0 0 0 0 1 0 ⎟
⎜ ⎟
⎜0 0 0 0 0 1 ⎟
⎝ ⎠
0 0 fp 0 fr fs
Then the rewritten differential equation is the system

U = AU + b with the initial condition
y x
′ ′′ ′
U (x, 0) = (ϕ(x), ϕ (x), ψ(x), ϕ (x), ψ (x), f0 (x)) , (3.5.25)
where f 0 (x)
′ ′′
= f (x, 0, ϕ(x), ϕ (x), ψ(x), ϕ (x), ψ (x))
′
.



3.5.1 Appendix: Real Analytic Functions
Multi-index notation
The following multi-index notation simplifies many presentations of formulas. Let x = (x 1, … , xn ) and
1 m
u : Ω ⊂ R ↦ R (or R for systems). (3.5.1 Appendix.1)
The n-tuple of non-negative integers (including zero)

α = (α1 , … , αn ) (3.5.1 Appendix.2)
is called multi-index. Set

|α| = α1 + … + αn
α! = α1 ! α2 ! ⋅ … ⋅ αn !
α α1 α2 αn
x = x x ⋅ … ⋅ xn (for a monom)
1 2
∂
Dk =
∂xk
D = (D1 , … , Dn )
Du = (D1 u, … , Dn u) ≡ ∇u ≡ grad u
|α|
α α1 α2 αn
∂
D = D D ⋅ … ⋅ Dn ≡ .
1 2 α1 α2 αn
∂x ∂x … ∂ xn
1 2
Define a partial order by

α ≥ β if and only if αi ≥ βi for all i. (3.5.1 Appendix.3)
Sometimes we use the notations

0 = (0, 0 … , 0), 1 = (1, 1 … , 1), (3.5.1 Appendix.4)
where
0, 1 ∈ R.
Using this multi-index notion, we have

1.
α!
α β γ
(x + y ) = ∑ x y , (3.5.1 Appendix.5)
β!γ!
β,γ
β+γ=α
where x, y ∈ R and α, β, γ are multi-indices.

2. Taylor expansion for a polynomial f (x) of degree m:
1
α α
f (x) = ∑ (D f (0)) x , (3.5.1 Appendix.6)
α!
|α|≤m
here is Dα
f (0) := (D
α
f (x)) |
x=0
.
3. Let x = (x 1, … , xn ) and m ≥ 0 an integer, then
$$
Erich Miersemann 3.5.1 Appendix.1 3/30/2022 https://math.libretexts.org/@go/page/2175

(x_1+\ldots+x_n)^m=\sum_{|\alpha|=m}\frac{m!}{\alpha!}x^\alpha.
\]
4.
$$
\alpha!\le|\alpha|!\le n^{|\alpha|}\alpha!.
\]
5. Leibniz's rule:
$$
D^\alpha(fg)=
\sum_{
β, γ
(3.5.1 Appendix.7)
β +γ = α
}\frac{\alpha!}{\beta!\gamma!}(D^\beta f)(D^\gamma g).

\]
6.
α!
β α α−β
D x = x if α ≥ β,
(α − β)!
β α
D x = 0 otherwise.
7. Directional derivative:
m
d |α|!
α α
f (x + ty) = ∑ (D f (x + ty)) y , (3.5.1 Appendix.8)
m
dt α!
|α|=m
where x, y ∈ R and t ∈ R . 1
8. Taylor's theorem: Let u ∈ C m+1

in a neighborhood N (y) of y , then, if x ∈ N (y),
1
α α
u(x) = ∑ (D u(y)) (x − y ) + Rm , (3.5.1 Appendix.9)
α!
|α|≤m
where
1 α α
Rm = ∑ (D u(y + δ(x − y))) x , 0 < δ < 1, (3.5.1 Appendix.10)
α!
|α|=m+1
δ = δ(u, m, x, y) ,
or
1
1 m (m+1)
Rm = ∫ (1 − t) Φ (t) dt, (3.5.1 Appendix.11)
m! 0
where Φ(t) = u(y + t(x − y)) . It follows from 7. that

$$
R_m=(m+1)\sum_{|\alpha|=m+1}\frac{1}{\alpha!}\left(\int_0^1\ (1-t)D^\alpha u(y+t(x-y))\ dt\right)(x-y)^\alpha.
\]
9. Using multi-index notation, the general linear partial differential equation of order m can be written as
$$

\sum_{|\alpha|\le m}a_\alpha (x)D^\alpha u=f(x)\ \ \mbox{in}\ \Omega\subset\mathbb{R}.
\]
Power series
Here we collect some definitions and results for power series in R.
Definition. Let c α
1
∈ R (or ∈ R
m
) . The series
∞
⎛ ⎞
∑ cα ≡ ∑ ∑ cα (3.5.1 Appendix.12)
⎝ ⎠
α m=0 |α|=m
is said to be convergent if
∞
⎛ ⎞
∑ | cα | ≡ ∑ ∑ | cα | (3.5.1 Appendix.13)
⎝ ⎠
α m=0 |α|=m
is convergent.
Remark. According to the above definition, a convergent series is absolutely 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, then
n ∞
α αi
∑x = ∏ (∑ x )
i
α i=1 αi =0
1
=
(1 − x1 )(1 − x2 ) ⋅ … ⋅ (1 − xn )
1
= ,
1
(1 − x)
provided |x i| <1 is satisfied for each i.

11. Assume x ∈ R and |x 1| + | x2 | + … + | xn | < 1 , then
∞
|α|! |α|!
α α
∑ x = ∑ ∑ x
α! α!
α j=0 |α|=j
j
= ∑(x1 + … + xn )
j=0
1
= .
1 − (x1 + … + xn )
12. Let x ∈ R, |x i| <1 for all i, and β is a given multi-index. Then

α! 1
α−β β
∑ x = D
1
(α − β)! (1 − x)
α≥β
β!
=
1+β
(1 − x)
13. Let x ∈ R and |x 1| + … + | xn | < 1 . Then

|α|! 1
α−β β
∑ x = D
(α − β)! 1 − x1 − … − xn
α≥β
|β|!
= .
1+|β|
(1 − x1 − … − xn )
Consider the power series

α
∑ cα x (3.34)
and assume this series is convergent for a z ∈ R . Then, by definition,

α
μ := ∑ | cα || z | <∞ (3.5.1 Appendix.14)
and the series (3.34) is uniformly convergent for all x ∈ Q(z), where
$$
Q(z):\ \ |x_i|\le|z_i|\ \ \mbox{for all}\ \ i.
\]
Figure 3.5.1.1: Definition of D ∈ Q(z)

Thus the power series (3.34) defines a continuous function defined on Q(z), according to a theorem of Weierstrass.
The interior of Q(z) is not empty if and only if z ≠ 0 for all i, see Figure 3.5.1.1.
i
For given x in a fixed compact subset D of Q(z) there is a q, 0 < q < 1 , such that
| xi | ≤ q| zi | for all i. (3.5.1 Appendix.15)
Set
$$
f(x)=\sum_\alpha c_\alpha x^\alpha.
\]
Proposition A1. (i) In every compact subset D of Q(z) one has f ∈ C (D) and ∞
the formal differentiate series, that is ∑ D c x , is uniformly convergent on the closure of D and is equal to D
α
β
α
α β
f .}
(ii)
β −|β|
| D f (x)| ≤ M |β|! r in D, (3.5.1 Appendix.16)
where
μ
M = , r = (1 − q) min | zi |. (3.5.1 Appendix.17)
n
(1 − q) 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_\alpha=\frac{1}{\alpha!}D^\alpha f(0).
\]
Definition. Assume f is defined on a domain Ω ⊂R , then f is said to be {\it real analytic in y ∈ Ω } if there are 1
cα ∈ R and if
there is a neighborhood N (y) of y such that
α
f (x) = ∑ cα (x − y ) (3.5.1 Appendix.18)
for all x ∈ N (y), and the series converges (absolutely) for each x ∈ N (y).
A function f is called {\it real analytic in Ω} if it is real analytic for each y ∈ Ω .
We will write f ∈ C (Ω) in the case that f is real analytic in the domain Ω.
ω
A vector valued function f (x) = (f (x), … , f ) is called real analytic if each coordinate is real analytic.
1 m
Proposition A2. (i) Let f ∈ C

ω
(Ω) . Then f ∈ C
∞
(Ω) .}
(ii)
Assume f ∈ C
ω
(Ω) . Then for each y ∈ Ω there exists a neighborhood N (y) and positive constants M , r such that
1
α α
f (x) = ∑ (D f (y))(x − y ) (3.5.1 Appendix.19)
α!
α
for all x ∈ N (y), and the series converges (absolutely) for each x ∈ N (y), and
β −|β|
| D f (x)| ≤ M |β|! r . (3.5.1 Appendix.20)
The proof follows from Proposition A1.

An open set Ω ∈ R is called connected if Ω is not a union of two nonempty
open sets with empty intersection. An open set Ω ∈ R is connected if and only if its path connected, see [11], pp. 38, for example.
We say that Ω is path connected if for any x, y ∈ Ω there is a continuous curve γ(t) ∈ Ω , 0 ≤ t ≤ 1 , with γ(0) = x and γ(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 (Ω) and Ω is connected. Then
ω
f is uniquely determined if for one z ∈ Ω all D f (z) are known.

α
Proof. See F. John [10], p. 65. Suppose g, h ∈ C (Ω) and ω
D g(z) = D h(z) for every α . Set f = g − h and

α α
α
Ω1 = {x ∈ Ω : D f (x) = 0 for all α},
α
Ω2 = {x ∈ Ω : D f (x) ≠ 0 for at least one α}.
The set Ω is open since

2 D
α
f are continuous in Ω . The set Ω1 is also open since f (x) = 0 in a neighbourhood of y ∈ Ω1 . This
follows from
1
α α
f (x) = ∑ (D f (y))(x − y ) . (3.5.1 Appendix.21)
α!
α
Since z ∈ Ω , i. e., Ω
1 1 ≠∅ , it follows Ω 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 satisfied.
∞

Definition. Let y ∈ Ω 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
β −|β|
| D f (y)| ≤ M |β|! r (3.5.1 Appendix.22)
for all β.
Proposition A4. f ∈ C
ω
(Ω) if and only if f ∈ C
∞
(Ω) and for every compact subset S ⊂Ω there are positive constants M, r
such that
f ∈ CM,r (y) for all y ∈ S. (3.5.1 Appendix.23)
Proof. See F. John [10], pp. 65-66. We will prove the local version of the proposition, that is, we show it for each fixed y ∈ Ω . The
general version follows from Heine-Borel theorem. Because of Proposition A3 it remains to show that the Taylor series
1 α α
∑ D f (y)(x − y ) (3.5.1 Appendix.24)
α!
α
converges (absolutely) in a neighborhood of y and that this series is equal to f (x).

Define a neighborhood of y by
Nd (y) = {x ∈ Ω : | x1 − y1 | + … + | xn − yn | < d}, (3.5.1 Appendix.25)
where d is a sufficiently small positive constant. Set Φ(t) = f (y + t(x − y)) . The one-dimensional Taylor theorem says
j−1
1 (k)
f (x) = Φ(1) = ∑ Φ (0) + rj , (3.5.1 Appendix.26)
k!
k=0
where
1
1 j−1 (j)
rj = ∫ (1 − t) Φ (t) dt. (3.5.1 Appendix.27)
(j − 1)! 0
From formula 7. for directional derivatives it follows for x ∈ N d (y) that

j
1 d 1 α α
Φ(t) = ∑ D f (y + t(x − y))(x − y ) . (3.5.1 Appendix.28)
j
j! dt α!
|α|=j
From the assumption and the multinomial formula 3. we get for 0 ≤ t ≤ 1

j
∣ 1 d ∣ |α|!
−|α| α
∣ Φ(t)∣ ≤ M ∑ r |(x − y ) |
j
∣ j! dt ∣ α!
|α|=j
−j j
= Mr (| x1 − y1 | + … + | xn − yn |)
j
d
≤ M( ) .
r
Choose d > 0 such that d < r , then the Taylor series converges (absolutely) in N (y) and it is equal to d f (x) since the remainder
satisfies, see the above estimate,
$$
|r_j|=\left|\frac{1}{(j-1)!}\int_0^1\ (1-t)^{j-1}\Phi^j(t)\ dt\right|\le M\left(\frac{d}{r}\right)^j.
\]

□
We recall that the notation f << F (f is majorized by F ) was defined in the previous section.
Proposition A5. (i) f = (f1 , … , fm ) ∈ CM,r (0) if and only if f << (Φ, … , Φ) , where
Mr
Φ(x) = . (3.5.1 Appendix.29)
r − x1 − … − xn
}
(ii) $f\in C_{M,r}(0)$ and f (0) = 0 if and only if
f << (Φ − M , … , Φ − M ), (3.5.1 Appendix.30)
where
M (x1 + … + xn )
Φ(x) = . (3.5.1 Appendix.31)
r − x1 − … − xn
Proof.
$$
D^\alpha\Phi(0)=M|\alpha|!r^{-|\alpha|}.
\]
□
Remark. The definition of f << F implies, trivially, that D α

f << D
α
F .
The next proposition shows that compositions majorize if the involved functions majorize. More precisely, we have
Proposition A6. Let f , F : R ↦ R
m
and maps a neighborhood of 0 ∈ R
g, G
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)). (3.5.1 Appendix.32)
For each coordinate h of h we have, according to the chain rule,

k
α β γ
D hk (0) = Pα (δ gl (0), D fj (0)), (3.5.1 Appendix.33)
where P are polynomials with non-negative integers as coefficients, P

α α
are independent on g or f and δ := (∂/∂ u , … , ∂/∂ u ) . Thus,

1 m
α β γ
|D hk (0)| ≤ Pα (| δ gl (0)|, | D fj (0)|)
β γ
≤ Pα (δ Gl (0), D Fj (0))
α
= D Hk (0).
Using this result and Proposition A4, which characterizes real analytic functions, 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 definition of g .
Proof. See F. John [10], p. 68. Assume that f maps a neighborhood of y ∈ R in R
m
and g maps a neighborhood of v = f (y) in
${\mathbb R}^m$. Then f ∈ C (y) and g ∈ C (v) implies
M,r μ,ρ
h(x) := g(f (x)) ∈ Cμ,ρr/(mM+ρ) (y). (3.5.1 Appendix.34)

Once one has shown this inclusion, the proposition follows from Proposition~A4. To show the inclusion, we set
∗ ∗
h(y + x) := g(f (y + x)) ≡ g(v + f (y + x) − f (x)) =: g (f (x)), (3.5.1 Appendix.35)
where v = f (y) and

∗
g (u) : = g(v + u) ∈ Cμ,ρ (0)
∗
f (x) : = f (y + x) − f (y) ∈ CM,r (0).
In the above formulas v, y are considered as fixed parameters. From Proposition~A5 it follows
∗
f (x) << (Φ − M , … , Φ − M ) =: F
∗
g (u) << (Ψ, … , Ψ) =: G,
where
Mr
Φ(x) =
r − x1 − x2 − … − xn
μρ
Ψ(u) = .
ρ − x1 − x2 − … − xn
From Proposition A6 we get

h(y + x) << (χ(x), … , χ(x)) ≡ G(F ), (3.5.1 Appendix.36)
where
μρ
χ(x) =
ρ − m(Φ(x) − M )
μρ(r − x1 − … − xn )
=
ρr − (ρ + mM )(x1 + … + xn )
μρr
<<
ρr − (ρ + mM )(x1 + … + xn )
μρr/(ρ + mM )
= .
ρr/(ρ + mM ) − (x1 + … xn )
See an exercise for the ''<<''-inequality.

□
Contributors:


3.E: Classification (Exercises)
calculus sequence.
Q3.1
Let χ : n
R
1
→ R in C
1
, ∇χ ≠ 0. Show that for given x0 ∈ R
n
there is in a neighborhood of x0 a local diffeomorphism
λ = Φ(x) , Φ : (x 1, … , xn ) ↦ (λ , … , λ ) , such that λ
1 n n = χ(x) .
Q3.2
Show that the differential equation
a(x, y)uxx + 2b(x, y)uxy + c(x, y)uyy + lower order terms = 0 (3.E.1)
is elliptic if ac − b 2
>0 , parabolic if ac − b 2
=0 and hyperbolic if ac − b 2
<0 .
Q3.3
Show that in the hyperbolic case there exists a solution of ϕ x + μ1 ϕy = 0 , see equation (3.9), such that ∇ϕ ≠ 0 .
Hint: Consider an appropriate Cauchy initial value problem.
Q3.4
Show equation (3.4).
Q3.5
Find the type of
Lu := 2 uxx + 2 uxy + 2 uyy = 0 (3.E.2)
and transform this equation into an equation with vanishing mixed derivatives by using the orthogonal mapping (transform to
principal axis) x = U y, U orthogonal.
Q3.6
Determine the type of the following equation at (x, y) = (1, 1/2).
Lu := x uxx + 2y uxy + 2xy uyy = 0. (3.E.3)
Q3.7
Find all C -solutions of
2
uxx − 4 uxy + uyy = 0. (3.E.4)
Hint: Transform to principal axis and stretching of axis lead to the wave equation.
Q3.8
Oscillations of a beam are described by
1
wx − σt = 0
E
σx − ρwt = 0,
where σ stresses, w deflection of the beam and E, ρ are positive constants.

a. Determine the type of the system.
b. Transform the system into two uncoupled equations, that is, w, σ occur only in one equation, respectively.

c. Find non-zero solutions.
Q3.9
Find nontrivial solutions (∇χ ≠ 0 ) of the characteristic equation to
2
x uxx − uyy = f (x, y, u, ∇u), (3.E.5)
where f is given.
Q3.10
Determine the type of
uxx − x uyx + uyy + 3 ux = 2x, (3.E.6)
where u = u(x, y).
Q3.11
Transform equation
2
uxx + (1 − y )uxy = 0, (3.E.7)
u = u(x, y) , into its normal form.
Q3.12
Transform the Tricomi-equation
y uxx + uyy = 0, (3.E.8)
u = u(x, y) , where y < 0 , into its normal form.
Q3.13
Transform equation
2 2
x uxx − y uyy = 0, (3.E.9)
u = u(x, y) , into its normal form.
Q3.14
Show that
1 1
λ = , Λ = . (3.E.10)
3/2 1/2
2 2
(1 + |p | ) (1 + |p | )
are the minimum and maximum of eigenvalues of the matrix (a ij

) , where
−1/2 pi pj
ij 2
a = (1 + |p | ) ( δij − ). (3.E.11)
2
1 + |p|
Q3.15
Show that Maxwell equations are a hyperbolic system.
Q3.16
Consider Maxwell equations and prove that div E = 0 and div H = 0 for all t if these equations are satisfied for a fixed time t . 0
Hint. div rot A = 0 for each C -vector field A = (A

2
1, A2 , A3 ) .

Q3.17
Assume a characteristic surface S(t) in R is defined by χ(x, y, z, t) = const. such that χ
3
t =0 and χz ≠0 . Show that S(t) has a
nonparametric representation z = u(x, y, t) with u = 0 , that is S(t) is independent of t .
t
Q3.18
Prove formula (3.22) for the normal on a surface.
Q3.19
Prove formula (3.23) for the speed of the surface S(t) .
Q3.20
Write the Navier-Stokes system as a system of type (3.4.1).
Q3.21
Show that the following system (linear elasticity, stationary case of (3.4.1.1) in the two-dimensional case) is elliptic
μ△u + (λ + μ)\ grad(div u) + f = 0, (3.E.12)
where u = (u , u ) . The vector f

1 2 = (f1 , f2 ) is given and
λ, μ are positive constants.
Q3.22
Discuss the type of the following system in stationary gas dynamics (isentrop flow) in R . 2
2
ρu ux + ρvuy + a ρx = 0
2
ρu vx + ρvvy + a ρy = 0
ρ(ux + vy ) + u ρx + vρy = 0.
Here are (u, v) velocity vector, ρ density and

− −−−
a = √p (ρ) the sound velocity.
′
Q3.23
Show formula 7. (directional derivative).
Hint: Induction with respect to m.
Q3.24
Let y = y(x) be the solution of:
′
y (x) = f (x, y(x))
y(x0 ) = y0 ,
where f is real analytic in a neighborhood of (x 0, y0 ) ∈ R

2
.
Find the polynomial P of degree 2 such that
3
y(x) = P (x − x0 ) + O(|x − x0 | ) (3.E.13)
as x → x . 0

Q3.25
Let u be the solution of
△u = 1
u(x, 0) = uy (x, 0) = 0.
Find the polynomial P of degree 2 such that

2 2 3/2
u(x, y) = P (x, y) + O((x +y ) ) (3.E.14)
as (x, y) → (0, 0).
Q3.26
Solve the Cauchy initial value problem
Mr
Vt = (1 + N (n − 1)Vs )
r −s−NV
V (s, 0) = 0.
Hint: Multiply the differential equation with (r − s − N V ) .
Q3.27
Write △ 2
u = −u as a system of first order.
Hint: △ 2
u ≡ △(△u) .
Q3.28
Write the minimal surface equation
⎛ ⎞ ⎛ ⎞
∂ ux ∂ uy
⎜ −−−−−−−−−⎟+ ⎜ −−−−−−−−−⎟ =0 (3.E.15)
∂x 2 2 ∂y 2 2
⎝ √ 1 + ux + uy ⎠ ⎝ √ 1 + ux + uy ⎠
as a system of first order.

−−−−−−−−− −−−−−−−−−
Hint: v 1
2 2
:= ux / √1 + ux + uy , v2 := uy / √1 + ux + uy .
2 2
Q3.29
Let f : R
1 m
×R → R
m
be real analytic in (x 0, y0 ) . Show that a real analytic solution in a neighborhood of x of the problem
0
′
y (x) = f (x, y)
y(x0 ) = y0
exists and is equal to the unique C 1

[ x0 − ϵ, x0 + ϵ] -solution from the Picard-Lindel\"of theorem, ϵ > 0 sufficiently small.



Back Matter
Index
Erich Miersemann 1 3/30/2022 https://math.libretexts.org/@go/page/44994

Index
B F L
Bessel's differential equation Fourier's Method Laplace equation
4.4: A Method of Riemann 6.4.1: Fourier's Method 1.3: Partial Differential Equations
C G P
CAPILLARY EQUATION Green's Function Parabolic Equations
1.3: Partial Differential Equations 7.4: Green's Function for $\Delta$ 6: Parabolic Equations
Poisson's formula
D I 6.1: Poisson's Formula
Diffusion in a tube Inhomogeneous Equations
6.4.1: Fourier's Method 4.3: Inhomogeneous Equations
DIRICHLET INTEGRAL
CHAPTER OVERVIEW
4: HYPERBOLIC EQUATIONS
Here we consider hyperbolic equations of second order, mainly wave equations.
4.1: ONE-DIMENSIONAL WAVE EQUATION

4.2: HIGHER DIMENSIONS
4.2.1: CASE N=3
4.2.2: CASE N=2
4.3: INHOMOGENEOUS EQUATIONS
4.4: A METHOD OF RIEMANN
Riemann's method provides a formula for the solution of the following Cauchy initial value problem for a hyperbolic equation of second
order in two variables

4.5.1: OSCILLATION OF A STRING
4.5.2: OSCILLATION OF A MEMBRANE
4.5.3: INHOMOGENEOUS WAVE EQUATIONS
4.E: 4: HYPERBOLIC EQUATIONS (EXERCISES)
BACK MATTER
INDEX
1 3/30/2022
CHAPTER OVERVIEW
FRONT MATTER
TITLEPAGE
INFOPAGE
1 3/30/2022
4: Hyperbolic Equations
integrated.

4.1: One-Dimensional Wave Equation
The one-dimensional wave equation is given by
1
utt − uxx = 0, (4.1.1)
2
c
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), (4.1.2)
with arbitrary C -functions f and g

2
The Cauchy initial value problem for the wave equation is to find a C -solution of 2
1
utt − uxx = 0
2
c
u(x, 0) = α(x)
ut (x, 0) = β(x),
where α, β ∈ C 2
(−∞, ∞) are given.
Theorem 4.1. There exists a unique C
2 1
(R ×R )
1
-solution of the Cauchy initial value problem, and this solution is given by
d'Alembert's1 formula
x+ct
α(x + ct) + α(x − ct) 1
u(x, t) = + ∫ β(s) ds. (4.1.3)
2 2c x−ct
Proof. Assume there is a solution u(x, t) of the Cauchy initial value problem, then it follows from (4.1.2) that
u(x, 0) = f (x) + g(x) = α(x) (4.1.4)
′ ′
ut (x, 0) = c f (x) − c g (x) = β(x). (4.1.5)
From (4.1.4) we obtain

′ ′ ′
f (x) + g (x) = α (x), (4.1.6)
which implies, together with (4.1.5), that

′
α (x) + β(x)/c
′
f (x) =
2
′
α (x) − β(x)/c
′
g (x) = .
2
Then
x
α(x) 1
f (x) = + ∫ β(s) ds + C1
2 2c 0
x
α(x) 1
g(x) = − ∫ β(s) ds + C2 .
2 2c 0
The constants C , C satisfy

1 2
C1 + C2 = f (x) + g(x) − α(x) = 0, (4.1.7)
see (4.1.4). Thus each C -solution of the Cauchy initial value problem is given by d'Alembert's formula. On the other hand, the
2
function u(x, t) defined by the right hand side of (4.1.3) is a solution of the initial value problem.
□

Corollaries. 1. The solution u(x, t) of the initial value problem depends on the values of α at the endpoints of the interval
[x − ct, x + ct] and on the values of β on this interval only, see Figure 4.1.1. The interval [x − ct, x + ct] is called {\it domain of
dependence}.
Figure 4.1.1: Interval of dependence

2. Let P be a point on the x-axis. Then we ask which points (x, t) need values of α or β at P in order to calculate u(x, t)? From
the d'Alembert formula it follows that this domain is a cone, see Figure 4.2.1. This set is called domain of influence.
Figure 4.2.1: Domain of influence

1
d'Alembert, Jean Babtiste le Rond, 1717-1783



4.2: Higher Dimensions
Set
$$
\Box u=u_{tt}-c^2\triangle u,\ \ \triangle\equiv\triangle_x=\partial^2/\partial x_1^2+\ldots+
\partial^2/\partial x_n^2,
\]
and consider the initial value problem
n 1
□u = 0 inR ×R (4.2.1)
u(x, 0) = f (x) (4.2.2)
ut (x, 0) = g(x), (4.2.3)
where f and g are given C 2 2

(R ) -functions.
By using spherical means and the above d'Alembert formula we will derive a formula for the solution of this initial value problem.
Method of Spherical means

Define the spherical mean for a C -solution u(x, t) of the initial value problem by
2
1
M (r, t) = ∫ u(y, t) dSy , (4.2.4)
ωn rn−1 ∂ Br (x)
where
$$
\omega_n=(2\pi)^{n/2}/\Gamma(n/2)
\]
is the area of the n-dimensional sphere, ω nr
n−1
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 M (r, t). (4.2.5)
r→0
Using the initial data, we have

1
M (r, 0) = ∫ f (y) dSy =: F (r) (4.2.6)
ωn rn−1 ∂ Br (x)
1
Mt (r, 0) = ∫ g(y) dSy =: G(r), (4.2.7)
n−1
ωn r ∂ Br (x)
which are the spherical means of f and g .

The next step is to derive a partial differential equation for the spherical mean. From definition (4.2.4) of the spherical mean we
obtain, after the mapping ξ = (y − x)/r , where x and r are fixed,
1
M (r, t) = ∫ u(x + rξ, t) dSξ . (4.2.8)
ωn ∂ B1 (0)
It follows

n
1
Mr (r, t) = ∫ ∑ uy (x + rξ, t)ξi dSξ
i
ωn ∂ B1 (0)
i=1
n
1
= ∫ ∑ uy (y, t)ξi dSy .
i
n−1
ωn r ∂ Br (x)
i=1
Integration by parts yields

n
1
∫ ∑ uy y (y, t) dy (4.2.9)
n−1 i i
ωn r Br (x)
i=1
since $\xi\equiv (y-x)/r$ is the exterior normal at ∂ B r (x) . Assume u is a solution of the wave equation, then
n−1
1
r Mr = ∫ utt (y, t) dy
2
c ωn Br (x)
r
1
= ∫ ∫ utt (y, t) dSy dc.
2
c ωn 0 ∂ Bc (x)
The previous equation follows by using spherical coordinates. Consequently

1
n−1
(r Mr )r = ∫ utt (y, t) dSy
2
c ωn ∂ Br (x)
n−1 2
r ∂ 1
= ( ∫ u(y, t) dSy )
2 2
c ∂t ωn rn−1 ∂ Br (x)
n−1
r
= Mtt .
2
c
Thus we arrive at the differential equation

n−1 −2 n−1
(r Mr )r = c r Mtt , (4.2.10)
which can be written as

n−1
−2
Mrr + Mr = c Mtt . (4.2.11)
r
This equation (4.2.11) 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). (4.2.1.1)
From the d'Alembert formula we get formally

(r + ct)F (r + ct) + (r − ct)F (r − ct)
M (r, t) = (4.2.1.1)
2r
r+ct
1
+ ∫ ξG(ξ) dξ. (4.2.1.2)
2cr r−ct
The right hand side of the previous formula is well defined if the domain of dependence [x − ct, x + ct] is a subset of (0, ∞). We
can extend F and G to F and G which are defined on (−∞, ∞) such that rF and rG are C (R ) -functions as follows.
0 0 0 0
2 1
Set
$$
F_0(r)=\left\{
F (r) r >0
f (x) r =0 (4.2.1.2)
F (−r) r <0
\right.\
\]
The function G 0 (r) is given by the same definition where F and f are replaced by G and g , respectively.
Lemma. rF 0 (r), rG0 (r) ∈ C
2
(R )
2
.
Proof. From definition of F (r) and G(r) , r > 0 , it follows from the mean value theorem
$$\lim_{r\to+0} F(r)=f(x),\ \ \ \lim_{r\to+0} G(r)=g(x).\]
Thus rF (r) and rG (r) are C (R )-functions. These functions are also in C
0 0
1 1
(R )
1
. This follows since F and G are in
0 0 C
1 1
(R ) .
We have, for example,
n
1
′
F (r) = ∫ ∑ fy (x + rξ)ξj dSξ
j
ωn ∂ B1 (0)
j=1
′
1
F (+0) = ∫ ∑ fy (x)ξj dSξ
j
ωn ∂ B1 (0) j=1
n
1
= ∑ fy (x) ∫ nj dSξ
j
ωn ∂ B1 (0)
j=1
= 0.
Then, rF 0 (r) and rG 0 (r) are in C 2

(R )
1
, provided F and G are bounded as r → +0 . This property follows from
′′ ′′
$$F''(r)=\dfrac{1}{\omega_n}\int_{\partial B_1(0)}\ \sum_{i,j=1}^n f_{y_iy_j}(x+r\xi)\xi_i\xi_j\ dS_\xi.\]

Thus
$$F''(+0)=\dfrac{1}{\omega_n}\sum_{i,j=1}^n f_{y_iy_j}(x)\int_{\partial B_1(0)}\ n_in_j\ dS_\xi.\]
We recall that f , g ∈ C 2
(R )
2
by assumption.
□

The solution of the above initial value problem, where F and G are replaced by F and G , respectively, is 0 0
(r + ct)F0 (r + ct) + (r − ct)F0 (r − ct)

M0 (r, t) =
2r
r+ct
1
+ ∫ ξ G0 (ξ) dξ.
2cr r−ct
Since F and G are even functions, we have

0 0
$$\int_{r-ct}^{ct-r}\ \xi G_0(\xi)\ d\xi=0.\]

Thus
(r + ct)F0 (r + ct) − (ct − r)F0 (ct − r)
M0 (r, t) =
2r
ct+r
1
+ ∫ ξ G0 (ξ) dξ, (4.2.1.3)
2cr ct−r
see Figure 4.2.1.1.
Figure 4.2.1.1: Changed domain of integration

For fixed t > 0 and 0 < r < ct it follows that M (r, t) is the solution of the initial value problem with given initially data (
0
4.2.1.1) since F (s) = F (s) , G (s) = G(s) if s > 0 .

0 0
Since for fixed t > 0

$$u(x,t)=\lim_{r\to 0} M_0(r,t),\]
it follows from d'Hospital's rule that
′
u(x, t) = ctF (ct) + F (ct) + tG(ct)
d
= (tF (ct)) + tG(ct).
dt
Theorem 4.2. Assume f ∈ C (R ) and g ∈ C (R ) are given. Then there exists a unique solution
3 3 2 3
u ∈ C
2 3
(R × [0, ∞)) of the
initial value problem (4.2.2)-(4.2.3), where n = 3 , and the solution is given by the Poisson's formula
1 ∂ 1
u(x, t) = ( ∫ f (y) dSy ) (4.2.1.3)
2
4πc ∂t t ∂ Bct (x)
1
+ ∫ g(y) dSy . (4.2.1.4)
2
4π c t ∂ Bct (x)
Proof. Above we have shown that a C -solution is given by Poisson's formula. Under the additional assumption f
2
∈ C
3
it follows
from Poisson's formula that this formula defines a solution which is in C , see F. John [10], p. 129.
2
Corollary. From Poisson's formula we see that the domain of dependence for u(x, t0 ) is the intersection of the cone defined by
|y − x| = c|t − t | with the hyperplane defined by t = 0 , see Figure 4.2.1.2.
0

Figure 4.2.1.2: Domain of dependence, case n = 3 .



4.2.2: Case n=2
Consider the initial value problem
−2
vxx + vyy = c vtt (4.2.2.1)
v(x, y, 0) = f (x, y) (4.2.2.2)
vt (x, y, 0) = g(x, y), (4.2.2.3)
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 Hadamard's method of decent.
Let v(x, y, t) be a solution of (4.2.2.1)-(4.2.2.3), 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 satisfies (
4.2.2.1)-(4.2.2.3). Hence, since u(x, y, z, t) = u(x, y, 0, t) + u (x, y, δz, t)z, 0 < δ < 1 , and u = 0 , we have
z z
$$v(x,y,t)=u(x,y,0,t).\]
Poisson's formula in the three-dimensional case implies
1 ∂ 1
v(x, y, t) = ( ∫ f (ξ, η) dS)
2
4πc ∂t t ∂ Bct (x,y,0)
1
+ ∫ g(ξ, η) dS. (4.2.2.4)
2
4π c t ∂ Bct (x,y,0)
Figure 4.2.2.1: Domains of integration

The integrands are independent on ζ . The surface S is defined by χ(ξ, η, ζ) := (ξ − x ) 2
+ (η − y )
2
+ζ
2 2
−c t . Then the
2
=0
exterior normal n at S is n = ∇χ/|∇χ| and the surface element is given by dS = (1/|n 3 |)dξdη , where the third coordinate of n
is
$$n_3=\pm\frac{\sqrt{c^2 t^2-(\xi-x)^2-(\eta-y)^2}}{ct}.\]
The positive sign applies on S
+
, where ζ >0 and the sign is negative on S
−
where ζ <0 , see Figure 4.2.2.1. We have
¯
¯¯¯¯¯
¯
S =S
+
∪S .−
−−−−−−−−−−−−−− −
Set ρ = √(ξ − x ) 2
+ (η − y )
2
. Then it follows from (4.2.2.4)
Theorem 4.3. The solution of the Cauchy initial value problem (4.2.2.1)-(4.2.2.3) is given by
1 ∂ f (ξ, η)
v(x, y, t) = ∫ − −−−−− − dξdη
2πc ∂t Bct (x,y) √ c2 t2 − ρ2
1 g(ξ, η)
+ ∫ dξdη.
−−−−−− −
2πc 2 2 2
Bct (x,y) √c t − ρ

Figure 4.2.2.2: Interval of dependence, case n = 2
Corollary. In contrast to the three dimensional case, the domain of dependence is here the disk B (x , y ) and not the boundary
cto 0 0
only. Therefore, see formula of Theorem 4.3, if f , g have supports in a compact domain D ⊂ R , then these functions have
2
influence on the value v(x, y, t) for all time t > T , T sufficiently large.



4.3: Inhomogeneous Equations
Here we consider the initial value problem
n 1
□u = w(x, t) on x ∈ R , t ∈ R (4.3.1)
u(x, 0) = f (x) (4.3.2)
ut (x, 0) = g(x), (4.3.3)
where □u := u tt
2
− c △u . We assume f ∈ C
3
, g ∈ C and w ∈ C , which are given.
2 1
Set u = u + u , where u is a solution of problem (4.3.1)-(4.3.3) with w := 0 and u is the solution where f
1 2 1 2 =0 and g = 0 in (
4.3.1)-(4.3.3). Since we have explicit solutions u in the cases n = 1 , n = 2 and n = 3 , it remains to solve
1
n 1
□u = w(x, t) on x ∈ R , t ∈ R (4.3.4)
u(x, 0) = 0 (4.3.5)
ut (x, 0) = 0. (4.3.6)
The following method is called Duhamel's principle which can be considered as a generalization of the method of variations of
constants in the theory of ordinary differential equations.
To solve this problem, we make the ansatz
t
u(x, t) = ∫ v(x, t, s) ds, (4.3.7)

0
where v is a function satisfying

□v = 0 for all s (4.3.8)
and
v(x, s, s) = 0. (4.3.9)
From ansatz (4.3.7) and assumption (4.3.9) we get

t
ut = v(x, t, t) + ∫ vt (x, t, s) ds,

0
= ∫ vt (x, t, s). (4.3.10)

0
It follows u (x, 0) = 0 . The initial condition

t u(x, t) = 0 is satisfied because of the ansatz (4.3.7). From (4.3.10) and ansatz (
4.3.7) we see that
utt = vt (x, t, t) + ∫ vtt (x, t, s) ds,

0
△x u = ∫ △x v(x, t, s) ds.

0
Therefore, since u is an ansatz for (4.3.4)-(4.3.6),

t
2
utt − c △x u = vt (x, t, t) + ∫ (□v)(x, t, s) ds
0
= w(x, t).
Thus necessarily v t (x, , see (4.3.8). We have seen that the ansatz provides a solution of (4.3.4)-(4.3.6) if for all s
t, t) = w(x, t)
□v = 0, v(x, s, s) = 0, vt (x, s, s) = w(x, s). (4.3.11)
Let v ∗
(x, t, s) be a solution of

□v = 0, v(x, 0, s) = 0, vt (x, 0, s) = w(x, s), (4.3.12)
then
$$v(x,t,s):=v^*(x,t-s,s)\]
is a solution of (4.3.11).
In the case n = 3 , where v is given by, see Theorem 4.2,
∗
$$v^*(x,t,s)=\frac{1}{4\pi c^2 t}\int_{\partial B_{ct}(x)}\ w(\xi,s)\ dS_\xi.\]

Then
∗
v(x, t, s) = v (x, t − s, s)
1
= ∫ w(ξ, s) dSξ .
2
4π c (t − s) ∂ Bc( t−s) (x)
from ansatz (4.3.7) it follows

t
u(x, t) = ∫ v(x, t, s) ds

0
t
1 w(ξ, s)
= ∫ ∫ dSξ ds.
2
4πc 0 ∂ Bc( t−s) (x)
t −s
Changing variables by τ = c(t − s) yields

ct
1 w(ξ, t − τ /c)
u(x, t) = ∫ ∫ dSξ dτ
2
4πc 0 ∂ Bτ (x)
τ
1 w(ξ, t − r/c)
= ∫ dξ,
2
4πc Bct (x)
r
where r = |x − ξ| .
Formulas for the cases n = 1 and n =2 follow from formulas for the associated homogeneous equation with inhomogeneous
initial values for these cases.
Theorem 4.4. The solution of
□u = w(x, t), u(x, 0) = 0, ut (x, 0) = 0, (4.3.13)
where w ∈ C , is given by:

1
Case n = 3 :
1 w(ξ, t − r/c)
u(x, t) = ∫ dξ, (4.3.14)
2
4πc Bct (x)
r
where r = |x − ξ| , x = (x 1, x2 , x3 ) , ξ = (ξ 1, ξ2 , ξ3 ) .
Case n = 2 :
t
1 w(ξ, τ )
u(x, t) = ∫ (∫ − −− − −−− − −−− − dξ) dτ , (4.3.15)
4πc 0 Bc( t−τ ) (x) √ c2 (t − τ )2 − r2
x = (x1 , x2 ) , ξ = (ξ
1, ξ2 ) .
Case n = 1 :
t x+c(t−τ)
1
u(x, t) = ∫ (∫ w(ξ, τ ) dξ) dτ . (4.3.16)
2c 0 x−c(t−τ)
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
Riemann's 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
$${\mathcal S}:\ \ x=x(t), y=y(t),\ \ t_1\le t \le t_2,\]
be a regular curve in R , that is, we assume x, y ∈ C
2 1
[ t1 , t2 ] 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 and c, f
1
∈ C in a neighborhood of S . Consider the initial value problem
Lu = f (x, y) (4.4.1)
u0 (t) = u(x(t), y(t)) (4.4.2)
p0 (t) = ux (x(t), y(t)) (4.4.3)
q0 (t) = uy (x(t), y(t)), (4.4.4)
where f ∈ C in a neighbourhood of S and u 0, p0 , q0 ∈ C

1
are given.
We assume:
i. u (t) = p (t)x (t) + q (t)y (t) (strip condition),
′
0 0
′
0
′
ii. S is not a characteristic curve. Moreover assume that the characteristic curves, which are lines here and are defined by
x = const. and y = const. , have at most one point of intersection with S , and such a point is not a touching point, i. e.,
tangents of the characteristic and S are different at this point.

We recall that the characteristic equation to (4.4.1) is χ χ = 0 which is satisfied if χ (x, y) = 0 or χ (x, y) = 0 . One family of
x y x y
characteristics associated to these first partial differential of first order is defined by x (t) = 1, y (t) = 0 , see Chapter 2.
′ ′
Assume u, v ∈ C and that u

1
xy , vxy exist and are continuous. Define the adjoint differential expression by
$$Mv=v_{xy}-(av)_x-(bv)_y+cv.\]
We have
2(vLu − uM v) = (ux v − vx u + 2buv)y + (uy v − vy u + 2auv)x . (4.4.5)
Set
P = −(ux v − xx u + 2buv)
Q = uy v − vy u + 2auv.
From (4.4.5) it follows for a domain Ω ∈ R 2
2∫ (vLu − uM v) dxdy = ∫ (−Py + Qx ) dxdy

Ω Ω
= ∮ P dx + Qdy, (4.4.6)
where integration in the line integral is anticlockwise. The previous equation follows from Gauss theorem or after integration by
parts:
$$\int_\Omega\ (-P_y+Q_x)\ dxdy=\int_{\partial\Omega}\ (-Pn_2+Qn_1)\ ds,\]
where n = (dy/ds, −dx/ds) , s arc length, (x(s), y(s)) represents ∂Ω .
Assume u is a solution of the initial value problem (4.4.1)-(4.4.4) and suppose that v satisfies
$$Mv=0\ \ \mbox{in}\ \Omega.\]

Figure 4.4.1: Riemann's method, domain of integration
Then, if we integrate over a domain Ω as shown in Figure 4.4.1, it follows from (4.4.6) that
2∫ vf dxdy = ∫ P dx + Qdy + ∫ P dx + Qdy + ∫ P dx + Qdy. (4.4.7)

Ω BA AP PB
The line integral from B to A is known from initial data, see the definition of P and Q.
Since
$$u_xv-v_xu+2buv=(uv)_x+2u(bv-v_x),\]
it follows
∫ P dx + Qdy = − ∫ ((uv)x + 2u(bv − vx )) dx

AP AP
= −(uv)(P ) + (uv)(A) − ∫ 2u(bv − vx ) dx.

AP
By the same reasoning we obtain for the third line integral
∫ P dx + Qdy = ∫ ((uv)y + 2u(av − vy )) dy

PB PB
= (uv)(B) − (uv)(P ) + ∫ 2u(av − vy ) dy.

PB
Combining these equations with (4.4.6), we get
2v(P )u(P ) = ∫ (ux v − vx + 2buv) dx − (uy v − vy u + 2auv) dy

BA
+u(A)v(A) + u(B)v(B) + 2 ∫ u(bv − vx ) dx

AP
+2 ∫ u(av − vy ) dy − 2 ∫ f v dxdy. (4.4.8)

PB Ω
Let v be a solution of the initial value problem, see Figure 4.2.2 for the definition of domain D(P ),
Figure 4.4.2: Definition of Riemann's function
M v = 0 in D(P ) (4.4.9)
bv − vx = 0 on C1 (4.4.10)
av − vy = 0 on C2 (4.4.11)
v(P ) = 1. (4.4.12)

Assume v satisfies (4.4.9)-(4.4.12), then
2u(P ) = u(A)v(A) + u(B)v(B) − 2 ∫ f v dxdy

Ω
=∫ (ux v − vx + 2buv) dx − (uy v − vy u + 2auv) dy,

BA
where the right hand side is known from given data.

A function v = v(x, y; x 0, y0 ) satisfying (4.4.9)-(4.4.12) is called Riemann's function.
Remark. Set w(x, y) = v(x, y; x 0, y0 ) for fixed x 0, y0 . Then (4.4.9)-(4.4.12) imply
x
w(x, y0 ) = exp( ∫ b(τ , y0 ) dτ ) on C1 ,

x0
w(x0 , y) = exp( ∫ a(x0 , τ ) dτ ) on C2 .

y0
Example 4.4.1:
uxy = f (x, y) , then a Riemann function is v(x, y) ≡ 1 .
Example 4.4.2:
Consider the telegraph equation of Chapter 3

$$\varepsilon \mu u_{tt}=c^2\triangle_xu-\lambda\mu u_t,\]
where u stands for one coordinate of electric or magnetic field.
Introducing
$$u=w(x,t)e^{\kappa t},\]
where κ = −λ/(2ε) , we arrive at
$$w_{tt}=\frac{c^2}{\varepsilon\mu}\triangle_xw-\frac{\lambda^2}{4\epsilon^2}.\]
Stretching the axis and transform the equation to the normal form we get finally 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.\]
−−
−
Substitution σ = √4cs leads to Bessel's differential equation
$$\sigma^2 z''(\sigma)+\sigma z'(\sigma)+\sigma^2 z(\sigma)=0,\]
where z(σ) = w(σ 2
/(4c)) . A solution is
$$J_0(\sigma)=J_0\left(\sqrt{4c(x-x_0)(y-y_0)}\right)\]
which defines a Riemann function since J 0 (0) =1 .
Remark. Bessel's differential equation is
$$x^2y''(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
1
bounded at 0. This bounded solution is the Bessel function J (x) of first kind and of order n , see [1], for example.
n



4.5: Initial-Boundary Value Problems
In previous sections we looked at solutions defined for all x ∈ R and t ∈ R . In this and in the following section we seek
n 1
solutions u(x, t) defined in a bounded domain Ω ⊂ R and for all t ∈ R and which satisfy additional boundary conditions on
n 1
∂Ω.



4.5.1: Oscillation of a String
Let u(x, t), x ∈ [a, b], t ∈ R , be the deflection of a string, see Figure 1.1.1 from Chapter 1. Assume the deflection occurs in the
1
(x, u)-plane. This problem is governed by the initial-boundary value problem
utt (x, t) = uxx (x, t) on (0, l) (4.5.1.1)
u(x, 0) = f (x) (4.5.1.2)
ut (x, 0) = g(x) (4.5.1.3)
u(0, t) = u(l, t) = 0. (4.5.1.4)
Assume the initial data f , g are sufficiently regular. This implies compatibility conditions f (0) = f (l) = 0 and g(0) = g(l) .
Fourier's method
To find solutions of differential equation (4.5.1.1) we make the separation of variables ansatz
u(x, t) = v(x)w(t). (4.5.1.1)
Inserting the ansatz into (4.5.1.1) we obtain

′′ ′′
v(x)w (t) = v (x)w(t), (4.5.1.2)
or, if v(x)w(t) ≠ 0 ,
′′ ′′
w (t) v (x)
= . (4.5.1.3)
w(t) v(x)
It follows, provided v(x)w(t) is a solution of differential equation (4.5.1.1) and v(x)w(t) ≠ 0 ,

′′
w (t)
= const. =: −λ (4.5.1.4)
w(t)
and
′′
v (x)
= −λ (4.5.1.5)
v(x)
since x, t are independent variables.

Assume v(0) = v(l) = 0 , then v(x)w(t) satisfies the boundary condition (4.5.1.4). Thus we look for solutions of the eigenvalue
problem
′′
−v (x) = λv(x) in (0, l) (4.5.1.5)
v(0) = v(l) = 0, (4.5.1.6)
which has the eigenvalues

2
π
λn = ( n) , n = 1, 2, … , (4.5.1.6)
l
and associated eigenfunctions are

π
vn = sin( nx). (4.5.1.7)
l

Solutions of
′′
−w (t) = λn w(t) (4.5.1.8)
are
−− −−
sin(√λn t), cos(√λn t). (4.5.1.9)
Set
−− −−
wn (t) = αn cos(√λn t) + βn sin(√λn t), (4.5.1.10)
where α , β ∈ R .
n n
1
It is easily seen that w (t)v

n n (x) is a solution of differential equation (4.5.1.1), and, since (4.5.1.1) is linear and homogeneous, also
(principle of superposition)
N
uN = ∑ wn (t)vn (x) (4.5.1.11)
n=1
which satisfies the differential equation (4.5.1.1) and the boundary conditions (4.5.1.4). Consider the formal solution of (4.5.1.1), (
4.5.1.4)
∞
−− −− −−
u(x, t) = ∑ (αn cos(√λn t) + βn sin(√λn t)) sin(√λn x). (4.5.1.7)
n=1
''Formal'' means that we know here neither that the right hand side converges nor that it is a solution of the initial-boundary value
problem. Formally, the unknown coefficients can be calculated from initial conditions (4.5.1.2), (4.5.1.3) as follows. We have
∞
−−
u(x, 0) = ∑ αn sin(√λn x) = f (x). (4.5.1.12)
n=1
−−
Multiplying this equation by sin(√λ k x) and integrate over (0, l), we get
l l
2
−− −−
αn ∫ sin (√λk x) dx = ∫ f (x) sin(√λk x) dx. (4.5.1.13)
0 0
We recall that
l
−− −− l
∫ sin(√λn x) sin(√λk x) dx = δnk . (4.5.1.14)
0
2
Then
l
2 πk
αk = ∫ f (x) sin( x) dx. (4.5.1.8)
l 0
l
By the same argument it follows from

∞
−− −−
ut (x, 0) = ∑ βn √λn sin(√λn x) = g(x) (4.5.1.15)
n=1

that
l
2 πk
βk = ∫ g(x) sin( x) dx. (4.5.1.9)
kπ 0
l
Under additional assumptions f ∈ C (0, l), g ∈ C (0, l) it follows that the right hand side of (4.5.1.7), where α , β are given by
4
0 0
3
n n
(4.5.1.8) and (4.5.1.9), respectively, defines a classical solution of (4.5.1.1)-(4.5.1.4) since under these assumptions the series for
u and the formal differentiate series for u , u , u , u
t tt x converges uniformly on 0 ≤ x ≤ l , 0 ≤ t ≤ T , 0 < T < ∞ fixed, see an
xx
exercise.



4.5.2: Oscillation of a Membrane
Let Ω ⊂ R be a bounded domain. We consider the initial-boundary value problem
2
1
utt (x, t) = △x u in Ω × R , (4.5.2.1)
¯
¯¯¯
u(x, 0) = f (x), x ∈ Ω, (4.5.2.2)
¯
¯¯¯
ut (x, 0) = g(x), x ∈ Ω, (4.5.2.3)
1
u(x, t) = 0 on ∂Ω × R . (4.5.2.4)
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 = λv in Ω, (4.5.2.5)
v = 0 on ∂Ω. (4.5.2.6)
Let λ are the eigenvalues of (4.5.2.5), (4.5.2.6) and v a complete associated orthonormal system of eigenfunctions. We assume
n n
Ω is sufficiently regular such that the eigenvalues are countable, which is satisfied in the following examples. Then the formal
solution of the above initial-boundary value problem is

$$
u(x,t)=\sum_{n=1}^\infty\left(\alpha_n\cos(\sqrt{\lambda_n}t)+\beta_n\sin(\sqrt{\lambda_n}t)\right)v_n(x),
\]
where
αn = ∫ f (x)vn (x) dx
Ω
1
βn = ∫ g(x)vn (x) dx.
−−
√λn Ω
Note
In general, eigenvalues of (4.5.2.5), (4.5.2.5) are not known explicitly. There are numerical methods to calculate these values.
In some special cases, see next examples, these values are known.
Examples
Example 4.5.2.1: Rectangle membrane
Let
Ω = (0, a) × (0, b). (4.5.2.1)
Using the method of separation of variables, we find all eigenvalues of (4.5.2.5), (4.5.2.6) which are given by
−−−−−−−
2 2
k l
λkl =√ + , k, l = 1, 2, … (4.5.2.2)
2 2
a b
and associated eigenfunctions, not normalized, are

$$
u_{kl}(x)=\sin\left(\frac{\pi k}{a}x_1\right)\sin\left(\frac{\pi l}{b}x_2\right).
\]

Example 4.5.2.2: Disk membrane
Set
2 2 2 2
Ω = {x ∈ R : x +x < R }. (4.5.2.3)
1 2
In polar coordinates, the eigenvalue problem (4.5.2.5), (4.5.2.6) is given by

1 1
− ((rur )r + uθθ ) = λu (4.5.2.6)
r r
u(R, θ) = 0, (4.5.2.7)
here is u = u(r, θ) := v(r cos θ, r sin θ) . We will find eigenvalues and eigenfunctions by separation of variables
u(r, θ) = v(r)q(θ), (4.5.2.4)
where v(R) = 0 and q(θ) is periodic with period 2π since u(r, θ) is single valued.
This leads to
1 1
′ ′ ′′
− ((rv ) q + vq ) = λvq. (4.5.2.5)
r r
Dividing by vq , provided vq ≠ 0 , we obtain

′ ′ ′′
1 (rv (r)) 1 q (θ)
− ( + ) = λ, (4.5.2.8)
r v(r) r q(θ)
which implies
′′
q (θ)
= const. =: −μ. (4.5.2.6)
q(θ)
Thus, we arrive at the eigenvalue problem

′′
−q (θ) = μq(θ)
q(θ) = q(θ + 2π).
It follows that eigenvalues μ are real and nonnegative. All solutions of the differential equation are given by
− −
q(θ) = A sin(√μ θ) + B cos(√μ θ), (4.5.2.7)
where A , B are arbitrary real constants. From the periodicity requirement

− − − −
A sin(√μ θ) + B cos(√μ θ) = A sin(√μ (θ + 2π)) + B cos(√μ (θ + 2π)) (4.5.2.8)
it follows
x +y x −y
sin x − sin y = 2 cos sin
2 2
x +y x −y
cos x − cos y = −2 sin sin
2 2

− − − − −
sin(√μ π) (A cos(√μ θ + √μ π) − B sin(√μ θ + √μ π)) = 0, (4.5.2.9)
which implies, since A , B are not zero simultaneously, because we are looking for q not identically zero,
− −
sin(√μ π) sin(√μ θ + δ) = 0 (4.5.2.10)
for all θ and a δ = δ(A, B, μ) . Consequently the eigenvalues are

2
μn = n , n = 0, 1, … . (4.5.2.11)
Inserting q ′′
(θ)/q(θ) = −n
2
into (4.5.2.8), we obtain the boundary value problem
2 ′′ ′ 2 2
r v (r) + rv (r) + (λ r − n )v = 0 on (0, R) (4.5.2.9)
v(R) = 0 (4.5.2.10)
sup |v(r)| < ∞. (4.5.2.11)

r∈(0,R)
− −
Set z = √λr and v(r) = v(z/√λ) =: y(z) , then, see (4.5.2.9),
2 ′′ ′ 2 2
z y (z) + zy (z) + (z − n )y(z) = 0, (4.5.2.12)
where z > 0 . Solutions of this differential equations which are bounded at zero are Bessel functions of first kind and n -th order
−
J (z) . The eigenvalues follows from boundary condition (4.5.2.10), i. e., from J (√λ R) = 0 . Denote by τ
n the zeros of
n nk
J (z) , then the eigenvalues of (4.5.2.6)-(4.5.2.6) are

n
2
τnk
λnk = ( ) (4.5.2.13)
R
and the associated eigenfunctions are

−−
−
Jn (√λnk r) sin(nθ), n = 1, 2, …
−−
−
Jn (√λnk r) cos(nθ), n = 0, 1, 2, … .
Thus the eigenvalues λ 0k are simple and λ nk , n ≥ 1 , are double eigenvalues.

Remark. For tables with zeros of Jn (x) and for much more properties of Bessel functions see \cite{Watson}. One has, in
particular, the asymptotic formula
1/2
2 1
Jn (x) = ( ) (cos(x − nπ/2 − π/5) + O ( )) (4.5.2.14)
πx x
as x → ∞ . It follows from this formula that there are infinitely many zeros of J n (x) .



4.5.3: Inhomogeneous Wave Equations
Let Ω ⊂ R be a bounded and sufficiently regular domain. In this section we consider the initial-boundary value problem
n
1
utt = Lu + f (x, t) in Ω × R (4.5.3.1)
¯
¯¯¯
u(x, 0) = ϕ(x) x ∈ Ω (4.5.3.2)
¯
¯¯¯
ut (x, 0) = ψ(x) x ∈ Ω (4.5.3.3)
1
u(x, t) = 0 for x ∈ ∂Ω and t ∈ R , (4.5.3.4)
where u = u(x, t), x = (x 1, ,

… , xn ) f , ϕ, ψ are given and L is an elliptic differential operator. Examples for L are:
1. L = ∂ /∂ x , oscillating string.
2 2
2. L = △ , oscillating membrane.
x
3. n
∂
ij
Lu = ∑ (a (x)ux ) , (4.5.3.1)
i
∂xj
i,j=1
¯
¯¯¯
where a = a are given sufficiently regular functions defined on Ω. We assume L is uniformly elliptic, that is, there is a constant
ij ji
ν > 0 such that
$$\sum_{i,j=1}^na^{ij}\zeta_i\zeta_j\ge\nu|\zeta|^2\]
for all x ∈ Ω and ζ ∈ R . n
4. Let u = (u 1, … , um ) and
$$Lu=\sum_{i,j=1}^n\frac{\partial}{\partial x_j}\left(A^{ij}(x)u_{x_i}\right),\]
¯
¯¯¯
where A = A are given sufficiently regular (m × m) -matrices on Ω. We assume that L defines an elliptic system. An example
ij ji
for this case is the linear elasticity.

Consider the eigenvalue problem
−Lv = λv in Ω (4.5.3.5)
v = 0 on ∂Ω. (4.5.3.6)
Assume there are infinitely many eigenvalues

$$0<\lambda_1\le\lambda_2\le\ldots\ \to\infty\]
and a system of associated eigenfunctions v , v 1 2, … which is complete and orthonormal in L 2
. This assumption is satisfied if
(Ω)
Ω is bounded and if ∂Ω is sufficiently regular.
For the solution of (4.5.3.1)-(4.5.3.4) we make the ansatz

∞
u(x, t) = ∑ vk (x)wk (t), (4.5.3.7)
k=1
with functions wk (t) which will be determined later. It is assumed that all series are convergent and that following calculations
make sense.
Let
∞
f (x, t) = ∑ ck (t)vk (x) (4.5.3.8)
k=1
be Fourier's decomposition of f with respect to the eigenfunctions v . We have k
ck (t) = ∫ f (x, t)vk (x) dx, (4.5.3.9)

Ω
which follows from (4.5.3.8) after multiplying with v (x) and integrating over Ω.
l
Set

$$\langle\phi,v_k\rangle=\int_\Omega\ \phi(x)v_k(x)\ dx,\]
then
∞
ϕ(x) = ∑⟨ϕ, vk ⟩vk (x)
k=1
ψ(x) = ∑⟨ψ, vk ⟩vk (x)
k=1
are Fourier's decomposition of ϕ and ψ , respectively.

In the following we will determine w k (t) , which occurs in ansatz (4.5.3.7), from the requirement that u = v k (x)wk (t) is a solution
of
$$u_{tt}=Lu+c_k(t)v_k(x)\]
and that the initial conditions
$$w_k(0)=\langle\phi,v_k\rangle,\ \ \ w_k'(0)=\langle\psi,v_k\rangle\]
are satisfied. From the above differential equation it follows
$$w_k''(t)=-\lambda_kw_k(t)+c_k(t).\]
Thus
−− −−
wk (t) = ak cos(√λk t) + bk sin(√λk t) (4.5.3.10)
t
1 −−
+ − − ∫ ck (τ ) sin(√λk (t − τ )) dτ ,
√λk 0
where
$$ a_k=\langle\phi,v_k\rangle,\ \ \ b_k=\frac{1}{\sqrt{\lambda_k}}\langle\psi,v_k\rangle.\]
Summarizing, we have
Proposition 4.2. The (formal) solution of the initial-boundary value problem (4.5.3.1)-(4.5.3.4) is given by
∞
u(x, t) = ∑ vk (x)wk (t), (4.5.3.2)
k=1
where v is a complete orthonormal system of eigenfunctions of (4.5.3.5), (4.5.3.6) and the functions w are defined by (4.5.3.10).
k k
The Resonance Phenomenon

Set in (4.5.3.1)-(4.5.3.4) ϕ = 0 , ψ = 0 and assume that the external force f is periodic and is given by
$$f(x,t)=A\sin(\omega t)v_n(x),\]
where A, ω are real constants and v is one of the eigenfunctions of (4.5.3.5), (4.5.3.6). It follows
n
$$c_k(t)=\int_\Omega\ f(x,t)v_k(x)\ dx=A\delta_{nk}\sin(\omega t).\]

Then the solution of the initial value problem (4.5.3.1)-(4.5.3.4) is
t
Avn (x) −−
u(x, t) = −− ∫ sin(ωτ ) sin(√λn (t − τ )) dτ
√λn 0
1 ω −−
= Avn (x) ( −− sin(√λk t) − sin(ωt)) ,
2
ω − λn √λn
−−
provided ω ≠ √λ . It follows
n
$$u(x,t)\to\frac{A}{2\sqrt{\lambda_n}}v_n(x)\left(\frac{\sin(\sqrt{\lambda_n}t)}{\sqrt{\lambda_n}}-t\cos(\sqrt{\lambda_n}
t)\right)\]

−− −−
if ω → √λ . The right hand side is also the solution of the initial-boundary value problem if ω = √λ .
n n
−− −−
Consequently |u| can be arbitrarily large at some points x and at some times t if ω = √λn . The frequencies √λn are called
critical frequencies at which resonance occurs.
A Uniqueness Result
¯
¯¯¯
The solution of the initial-boundary value problem (4.5.3.1)-(4.5.3.4) is unique in the class C 2
(Ω × R )
1
.
Proof. Let u , u are two solutions, then u = u
1 2 2 − u1 satisfies
1
utt = Lu in Ω × R
¯
¯¯¯
u(x, 0) = 0 x ∈ Ω
¯
¯¯¯
ut (x, 0) = 0 x ∈ Ω
n
u(x, t) = 0 for x ∈ ∂Ω and t ∈ R .
As an example we consider Example 3 from above and set

$$E(t)=\int_\Omega\ (\sum_{i,j=1}^na^{ij}(x)u_{x_i}u_{x_j}+u_tu_t)\ dx.\]
Then
n
′ ij
E (t) = 2 ∫ ( ∑ a (x)ux ux t + ut utt ) dx
i j
Ω
i,j=1
n
ij
= 2∫ ( ∑ a (x)uxi ut nj ) dS
∂Ω i,j=1
+2 ∫ ut (−Lu + ut t) dx

Ω
= 0.
It follows E(t) = const. From u (x, 0) = 0 and u(x, 0) = 0 we get E(0) = 0 . Consequently E(t) = 0 for all t , which implies,
t
¯
¯¯¯
since L is elliptic, that u(x, t) = const. on Ω × R . Finally, the homogeneous initial and boundary value conditions lead to
1
¯
¯¯¯
u(x, t) = 0 on Ω × R .1



4.E: 4: Hyperbolic Equations (Exercises)
2
utt = c uxx (4.E.1)
u(A) + u(C ) = u(B) + u(D) (4.E.2)
holds for all parallelograms ABC D in the (x, t)-plane, which are bounded by characteristic lines, see Figure 4.E.1.
Figure 4.1: Figure to the exercise
Q4.2: Method of separation of variables

Let vk (x) be an eigenfunction to the eigenvalue of the eigenvalue problem
′′
−v (x) = λv(x) in (0, l), v(0) = v(l) = 0
and let w k (t) be a solution of differential equation
′′
−w (t) = λk w(t) (4.E.3)
Prove that v k (x)wk (t) is a solution of the partial differential equation (wave equation) u tt = uxx .
Q4.3
Solve for given f (x) and μ ∈ R the initial value problem
1
1 1
ut + ux + μuxxx = 0 in R × R+
u(x, 0) = f (x) .
Q4.4
Let S := {(x, t); t = γx} be space-like, i.e., |γ| < 1/c ) in (x, t)-space, x = (x , x
2
1 2, x3 ) . Show that the Cauchy initial value
problem □u = 0 with data for u on S can be transformed using the Lorentz-transform
2
x1 − γ c t
x1 = (4.E.4)
− −−− − −−
√ 1 − γ 2 c2
′ ′
x = x2 , x = x3 (4.E.5)
2 3
t − γx1
′
t = (4.E.6)
−−−− −−−
2 2
√1 − γ c
into the initial value problem, in new coordinates,

□u = 0
′ ′
u(x , 0) = f (x )
′ ′
ut′ (x , 0) = g(x ) .
Here we denote the transformed function by u again.

Q4.5
∞
πn πn
u(x, t) := ∑ αn cos( t) sin( x) (4.E.7)
l l
n=1
is a C -solution of the wave equation u

2
tt = uxx if |αn| ≤ c/ n
4
, where the constant c is independent of n .
(ii) Set
l
πn
αn := ∫ f (x) sin( x) dx. (4.E.8)
0
l
Prove |α n| ≤ c/ n
4
, provided f ∈ C
0
4
(0, l) .
Q4.6
Let Ω be the rectangle (0, a) × (0, b) . Find all eigenvalues and associated eigenfunctions of −△u = λu in Ω, u = 0 on ∂Ω . Hint:
Separation of variables.
Q4.7
2
ℏ
n 1
iℏψt = − △x ψ + V (x)ψ in R ×R , (4.E.9)
2m
n
2
∫ |ψ(x, t)| dx = 1 , (4.E.10)
R
2m
n
△u + (E − V (x))u = 0 in R (4.E.11)
2
ℏ
n 2
under the side condition ∫ R
|u | dx = 1 , u : R n
↦ C .
Here is
n 1
ψ : R ×R ↦ C (4.E.12)
Planck's constant (ℏ ) is a small positive constant) and 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 = −me /(2ℏ n ) , n ∈ N , see [22], pp. 202.
n
4 2 2
Q4.8
Find nonzero solutions by using separation of variables of u = △ tt xu in Ω × (0, ∞) , u(x, t) = 0 on ∂Ω , where Ω is the circular
cylinder Ω = {(x , x , x ) ∈ R : x + x < R , 0 < x < h} .
1 2 3
n 2
1
2
2
2
3
Q4.9
3 utt − 4 uxx = 0
u(x, 0) = sin x
ut (x, 0) = 1 .
Q4.10
2 2 1
utt − c uxx = x , t > 0, x ∈ R
u(x, 0) = x
ut (x, 0) = 0 .
Hint: Find a solution of the differential equation independent on t , and transform the above problem into an initial value problem
with homogeneous differential equation by using this solution.

Q4.11
Find with the method of separation of variables nonzero solutions u(x, t), 0 ≤ x ≤ 1, 0 ≤ t < ∞, of
utt − uxx + u = 0 , (4.E.13)
such that u(0, t) = 0 , and u(1, t) = 0 for all t ∈ [0, ∞).
Q4.12
2 2
utt − c uxx = λ u, λ = const. (4.E.14)
2 2 2 2 2 2
u(x, t) = f (x − c t ) = f (s), s := x −c t (4.E.15)
2 ′′ ′ 2 2
z f (z) + zf (z) + (z − n )f = 0, z > 0 (4.E.16)
with n = 0 .
Remark. The above differential equation for u is the transformed telegraph equation (see Section 4.4).
Q4.13
Find the formula for the solution of the following Cauchy initial value problem uxy = f (x, y), where S : y = ax + b , a >0 , and
the initial conditions on S are given by
u = αx + βy + γ,
ux = α,
uy = β,
a, b, α, β, γ constants.
Q4.14
Find all eigenvalues μ of
′′
−q (θ) = μq(θ)
q(θ) = q(θ + 2π) .



Back Matter
Index

Index
B F L
C G P
Poisson's formula
DIRICHLET INTEGRAL
CHAPTER OVERVIEW
5: FOURIER TRANSFORM
Fourier's transform is an integral transform which can simplify investigations for linear differential or integral equations since it transforms a
differential operator into an algebraic equation.
Thumbnail: The real and imaginary parts of the Fourier transform of a delayed pulse. The Fourier transform decomposes a function into
eigenfunctions for the group of translations. (CC-BY-SA-4.0; Sławomir_Biały).

5.1: DEFINITION AND PROPERTIES

5.1.1: PSEUDODIFFERENTIAL OPERATORS
5.E: FOURIER TRANSFORM (EXERCISES)
1 3/30/2022
5.1: Definition and Properties
Definition. Let f ∈ C
0
s n
(R ) , s = 0, 1, … . The function f^ defined by
ˆ(ξ) = (2π )−n/2 ∫

f e
−iξ⋅x
f (x) dx, (5.1.1)
n
R
where ξ ∈ R , is called {\it Fourier transform} of f , and the function g̃

n
given by
−n/2 iξ⋅x
g̃ (x) = (2π ) ∫ e g(ξ) dξ (5.1.2)
n
R
is called inverse Fourier transform, provided the integrals on the right hand side
exist.
From (5.1.1) it follows by integration by parts that differentiation of a function 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.
α̂ |α| α ˆ
D f (ξ) = i ξ f (ξ), (5.1.3)
where |α| ≤ s .
The following proposition shows that the Fourier transform of f decreases rapidly for |ξ| → ∞ , provided f ∈ C
0
s n
(R ) . In
particular, the right hand side of (5.1.2) exists for g := f^ if f ∈ C n+1
0
(R ) .
n
Proposition 5.2. Assume g ∈ C (R ) , then there is a constant M

s
0
n
= M (n, s, g) such that
$$
|\widehat{g}(\xi)|\le \frac{M}{(1+|\xi|)^s}.
\]
Proof. Let ξ = (ξ , … , ξ
1 n) be fixed and let j be an index such that
| ξ | = max | ξ | . Then
j k k
1/2
n
2 −
|ξ| = ( ∑ ξ ) ≤ √n | ξj | (5.1.4)
k
k=1
which implies
s
s
s k
(1 + |ξ|) = ∑( )|ξ |
k
k=0
s
s k/2 k
≤ 2 ∑n | ξj |
k=0
s s/2 α
≤ 2 n ∑ |ξ |.
|α|≤s
This inequality and Proposition 5.1 imply

s s s/2 α
(1 + |ξ| ) | ĝ (ξ)| ≤ 2 n ∑ |(iξ ) ĝ (ξ)|
|α|≤s
s s/2 α
≤ 2 n ∑ ∫ | D g(x)| dx =: M .
n
R
|α|≤s
The notation inverse Fourier transform for (5.1.2) is justified by

˜ ˆ
Theorem 5.1. fˆ = f and f˜ = f .
Proof. See \cite{Yosida}, for example. We will prove the first assertion
−n/2 iξ⋅x ˆ
(2π ) ∫ e f (ξ) dξ = f (x) (5.1.5)
n
R
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
ˆ ix⋅ξ
∫ g(ξ)f (ξ)e dξ = ∫ ĝ (y)f (x + y) dy (5.1.6)
n n
R R
follows by direct calculation:
−n/2 −ix⋅y ix⋅ξ

∫ g(ξ) ((2π ) ∫ e f (y) dy) e dξ
n n
R R
−n/2 −iξ⋅(y−x)
= (2π ) ∫ (∫ g(ξ)e dξ) f (y) dy
n n
R R
=∫ ĝ (y − x)f (y) dy

n
R
=∫ ĝ (y)f (x + y) dy.

n
R
(ii) Formula
−n/2 −iy⋅ξ −n
(2π ) ∫ e g(εξ) dξ = ε ĝ (y/ε) (5.1.7)
n
R
for each ε > 0 follows after substitution z = εξ in the left hand side of (5.1.1).
(iii) Equation
ˆ ix⋅ξ
∫ g(εξ)f (ξ)e dξ = ∫ ĝ (y)f (x + εy) dy (5.1.8)
n n
R R
follows from (5.1.6) and (5.1.7). Set G(ξ) := g(εξ) , then (5.1.6) implies
ˆ ix⋅ξ ˆ
∫ G(ξ)f (ξ)e dξ = ∫ G(y)f (x + y) dy. (5.1.9)
n n
R R
Since, see (5.1.7),
ˆ −n/2 −iy⋅ξ
G(y) = (2π ) ∫ e g(εξ) dξ
n
R
−n
= ε ĝ (y/ε),

we arrive at
ˆ −n
∫ g(εξ)f (ξ) = ∫ ε ĝ (y/ε)f (x + y) dy
n n
R R
= ∫ ĝ (z)f (x + εz) dz.
n
R
Letting ε → 0 , we get
ˆ ix⋅ξ
g(0) ∫ f (ξ)e dξ = f (x) ∫ ĝ (y) dy. (5.1.10)
n n
R R
Set
2
−|x| /2
g(x) := e , (5.1.11)
then
n/2
∫ ĝ (y) dy = (2π ) . (5.1.12)
n
R
Since g(0) = 1 , the first assertion of Theorem 5.1 follows from (5.1.10) and (5.1.12). It remains to show (5.1.12).
(iv) Proof of (5.1.12). We will show
2
−n/2 −|x| /2 −ix⋅x
ĝ (y) : = (2π ) ∫ e e dx
n
R
2
−|y | /2
= e .
The proof of
2
−|y | /2 n/2
∫ e dy = (2π ) (5.1.13)
n
R
is left as an exercise. Since

2 2
x y x y |x| |y|
−( – +i –)⋅( – +i – ) = −( + ix ⋅ y − ) (5.1.14)
√2 √2 √2 √2 2 2
it follows
2 2 2
−|x| /2 −ix⋅y −η −|y | /2
∫ e e dx = ∫ e e dx
n n
R R
2 2
−|y | /2 −η
= e ∫ e dx
n
R
2 2
n/2 −|y | /2 −η
= 2 e ∫ e dη
n
R
where
x y
η := – +i –. (5.1.15)
√2 √2

Consider first the one-dimensional case. According to Cauchy's theorem we have
2
−η
∮ e dη = 0, (5.1.16)
C
where the integration is along the curve C which is the union of four curves as indicated in Figure ??? .
Figure 5.1.1: Proof of (5.1.12)

Consequently
R
−η
2 1 2
−x /2 −η
2
−η
2
∫ e dη = – ∫ e dx − ∫ e dη − ∫ e dη. (5.1.17)

C3 √2 −R C2 C4
It follows
−η
2
−
lim ∫ e dη = √π (5.1.18)
R→∞
C3
since
2
−η
lim ∫ e dη = 0, k = 2, 4. (5.1.19)
R→∞
Ck
The case n > 1 can be reduced to the one-dimensional case as follows. Set
x y
η = – +i – = (η1 , … , ηn ), (5.1.20)
√2 √2
where
xl yl
ηl = – +i –. (5.1.21)
√2 √2
From dη = dη 1 … dηl and

n
2 n
2 2
−η − ∑l=1 η −η
e =e l
= ∏e l
(5.1.22)
l=1
it follows
n
2 2
−eta −η
∫ e dη = ∏ ∫ e l
dηl , (5.1.23)
n
R Γl
l=1
where for fixed y

$$

\Gamma_l=\{z\in{\mathbb C}:\ z=\frac{x_l}{\sqrt{2}}+i\frac{y_l}{\sqrt{2}}, -\infty<x_l<+\infty\}.
\]
□
There is a useful class of functions for which the integrals in the definition of fˆ and f˜ exist.
For u ∈ C (R ) we set
∞ n
$$
q_{j,k}(u):=\max_{\alpha:\ |\alpha|\le k}\left(\sup_{\mathbb{R}^n}\left((1+|x|^2)^{j/2}|D^\alpha u(x)|\right)\right).
\]
Definition. The Schwartz class of rapidly decreasing functions is
$$
{\mathcal{S}}(\mathbb{R}^n)=\left\{u\in C^\infty(\mathbb{R}^n): \ q_{j,k}(u)<\infty\ \mbox{for any}\ j,k\in{\mathbb N}\cup\
{0\}\right\}.
\]
This space is a Frechét space.
Proposition 5.3. Assume u ∈ S(R ), then û and ũ ∈ S(R
n n
) .
Proof. See [24], Chapter 1.2, for example, or an exercise.



5.1.1: Pseudodifferential Operators
The properties of Fourier transform lead to a general theory for linear partial differential or integral equations. In this subsection we
define
$$D_k=\frac{1}{i}\frac{\partial}{\partial x_k},\ \ k=1,\ldots,n,\]
and for each multi-index α as in Subsection 3.5.1
$$D^\alpha=D_1^{\alpha_1}\ldots D_n^{\alpha_n}.\]
Thus
$$D^\alpha=\frac{1}{i^{|\alpha|}}\frac{\partial^{|\alpha|}}{\partial x_1^{\alpha_1}\ldots \partial x_n^{\alpha_n}}.\]
Let
$$p(x,D):=\sum_{|\alpha|\le m}a_\alpha(x)D^\alpha,\]
be a linear partial differential of order m, where a are given sufficiently regular functions.
α
According to Theorem 5.1 and Proposition 5.3, we have, at least for u ∈ S(R ), n
$$u(x)=(2\pi)^{-n/2}\int_{\mathbb{R}^n}\ e^{ix\cdot\xi}\widehat{u}(\xi)\ d\xi,\]

which implies
$$D^\alpha u(x)=(2\pi)^{-n/2}\int_{\mathbb{R}^n}\ e^{ix\cdot\xi}\xi^\alpha\widehat{u}(\xi)\ d\xi.\]
Consequently
−n/2 ix⋅ξ
p(x, D)u(x) = (2π ) ∫ e p(x, ξ)û (ξ) dξ, (5.1.1.1)
n
R
where
$$p(x,\xi)=\sum_{|\alpha|\le m}a_\alpha(x)\xi^\alpha.\]
The right hand side of (5.1.1.1) makes sense also for more general functions p(x, ξ), not only for polynomials.
Definition. The function p(x, ξ) is called symbol and
$$(Pu)(x):=(2\pi)^{-n/2}\int_{\mathbb{R}^n}\ e^{ix\cdot\xi}p(x,\xi)\widehat{u}(\xi)\ d\xi\]
is said to be pseudodifferential operator.
An important class of symbols for which the right hand side in this definition of a pseudodifferential operator is defined is S
m
which is the subset of p(x, ξ) ∈ C (Ω × R ) such that

∞ n
$$|D^\beta_xD_\xi^\alpha p(x,\xi)|\le C_{K,\alpha,\beta}(p)\left(1+|\xi|\right)^{m-|\alpha|}\]

for each compact K ⊂ Ω .
Above we have seen that linear differential operators define a class of pseudodifferential operators. Even integral operators can be
written (formally) as pseudodifferential operators.
Let
$$(Pu)(x)=\int_{\mathbb{R}^n}\ K(x,y)u(y)\ dy\]
be an integral operator. Then
−n/2 ix⋅ξ α
(P u)(x) = (2π ) ∫ K(x, y) ∫ e ξ û (ξ) dξ
n n
R R
−n/2 ix⋅ξ i(y−x)⋅ξ

= (2π ) ∫ e (∫ e K(x, y) dy) û (ξ).
n n
R R
Then the symbol associated to the above integral operator is

$$p(x,\xi)=\int_{\mathbb{R}^n}\ e^{i(y-x)\cdot\xi}K(x,y)\ dy.\]



5.E: Fourier Transform (Exercises)
calculus sequence.
Q5.1
Show
2
−|y | /2 n/2
∫ e dy = (2π ) . (5.E.1)
n
R
Q5.2
Show that u ∈ S(Rn implies u
)
^, ũ ∈ S(R
n
) .
Q5.3
Give examples for functions p(x, ξ) which satisfy p(x, ξ) ∈ S . m
Q5.4
Find a formal solution of Cauchy's initial value problem for the wave equation by using Fourier's transform.
Contributors:


CHAPTER OVERVIEW
6: PARABOLIC EQUATIONS
Here we consider linear parabolic equations of second order.
6.1: POISSON'S FORMULA

6.2: INHOMOGENEOUS HEAT EQUATION
6.3: MAXIMUM PRINCIPLE
6.4.1: FOURIER'S METHOD
6.4.2: UNIQUENESS
6.5: BLACK-SCHOLES EQUATION
Page notifications Off Donate Solutions of the Black-Scholes equation define the value of a derivative, for example of a call or put option,
which is based on an asset. An asset can be a stock or a derivative of it, for instance. In principle, there are infinitely many such products,
for example n-th derivatives.
6.E: PARABOLIC EQUATIONS (EXERCISES)
1 3/30/2022
6.1: Poisson's Formula
Assume u is a solution of (6.2), then, since Fourier transform is a linear mapping,
ˆ
ut − △u = ^
0. (6.1.1)
From properties of the Fourier transform, see Proposition 5.1, we have

n ˆ2 n
∂ u 2
ˆ 2
△u = ∑ = ∑ i ξ û (ξ), (6.1.2)
2 k
∂x
k=1 k k=1
provided the transforms exist. Thus we arrive at the ordinary differential equation for the Fourier transform of u
dû 2
+ |ξ | û = 0, (6.1.3)
dt
where ξ is considered as a parameter. The solution is

2
−|ξ| t
û (ξ, t) = ϕ̂ (ξ)e (6.1.4)
since û(ξ, 0) = ϕ
ˆ
(ξ) . From Theorem 5.1 it follows
2
−n/2 −|ξ| t iξ⋅x
u(x, t) = (2π ) ∫ ϕ̂ (ξ)e e dξ
n
R
2
−n iξ⋅(x−y)−|ξ| t
= (2π ) ∫ ϕ(y) ( ∫ e dξ) dy.
n n
R R
Set
$$K(x,y,t)=(2\pi)^{-n}\int_{\mathbb{R}^n}e^{i\xi\cdot (x-y)-|\xi|^2t}\ d\xi.\]
By the same calculations as in the proof of Theorem 5.1, step (vi), we find
2
−n/2 −|x−y | /4t
K(x, y, t) = (4πt) e . (6.1.5)
Figure 6.1.1: Kernel K(x, y, t), ρ = |x − y| , t 1 < t2
Thus we have
1 2
−|x−z| /4t
u(x, t) = ∫ ϕ(z)e dz. (6.1.6)
−
− n
(2 √πt)
n
R
Definition. Formula (6.1.6) is called Poisson's formula} and the function K defined by (6.1.5) is called heat kernel or
fundamental solution of the heat equation.
Proposition 6.1 The kernel K has following properties:
1. (i) K(x, y, t) ∈ C (R × R × R ) ,
∞ n n 1
+
2. (ii) (∂/∂t − △)K(x, y, t) = 0, t > 0 ,

3. (iii) K(x, y, t) > 0, t > 0 ,
4. (iv) ∫ K(x, y, t) dy = 1 , x ∈ R , t > 0
R
n
n
5. δ > 0 :(v) For each fixed

lim ∫ K(x, y, t) dy = 0 (6.1.7)
n
t→0 R ∖ Bδ (x)
t>0
uniformly for x ∈ R.
Proof. (i) and (iii) are obviously, and (ii) follows from the definition of K . Equations (iv) and (v) hold since
2
−n/2 −|x−y | /4t
∫ K(x, y, t) dy = ∫ (4πt) e dy
n n
R ∖ Bδ (x) R ∖ Bδ (x)
2
−n/2 −|η|
= π ∫ e dη
n
R ∖ Bδ/√4t (0)
by using the substitution y = x + (4t) 1/2

η . For fixed δ > 0 it follows (v) and for δ := 0 we obtain (iv).
□
Theorem 6.1. Assume ϕ ∈ C (R ) and sup |ϕ(x)| < ∞ . Then u(x, t) given by Poisson's formula (6.1.6) is in
n
R
n C
∞ n
(R ×R
1
+
) ,
continuous on R × [0, ∞) and a solution of the initial value problem (6.2), (6.3).
n
Proof. It remains to show

$$
\lim_{
x → ξ
(6.1.8)
t → 0
}u(x,t)=\phi(\xi).
\]
Figure 6.1.2: Figure to the proof of Theorem 6.1

Since ϕ is continuous there exists for given ε > 0 a δ = δ(ε) such that |ϕ(y) − ϕ(ξ)| < ε if |y − ξ| < 2δ .
Set M := sup |ϕ(y)| . Then, see Proposition 6.1,
R
n
u(x, t) − ϕ(ξ) = ∫ K(x, y, t) (ϕ(y) − ϕ(ξ)) dy. (6.1.9)

n
R
It follows, if |x − ξ| < δ and t > 0 , that
|u(x, t) − ϕ(ξ)| ≤ ∫ K(x, y, t) |ϕ(y) − ϕ(ξ)| dy

Bδ (x)
+∫ K(x, y, t) |ϕ(y) − ϕ(ξ)| dy

n
R ∖ Bδ (x)
≤ ∫ K(x, y, t) |ϕ(y) − ϕ(ξ)| dy

B2δ (x)
+2M ∫ K(x, y, t) dy

n
R ∖ Bδ (x)
≤ ε∫ K(x, y, t) dy + 2M ∫ K(x, y, t) dy

n n
R R ∖ Bδ (x)
< 2ε

if 0 < t ≤ t , t sufficiently small.
0 0
Remarks. 1. Uniqueness follows under the additional growth assumption

2
a|x|
|u(x, t)| ≤ M e in DT , (6.1.10)
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 , see [10], pp. 222.
T
2. u(x, t) defined by Poisson's formula depends on all values ϕ(y), y ∈ R . That means, a perturbation of ϕ , even far from a fixed
n
x, has influence to the value u(x, t). This means that heat travels with infinite 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+ ) ,
n
ut − △u = f (x, t) in x ∈ R , t ≥ 0,
u(x, 0) = ϕ(x),
where ϕ and f are given. From

$$ \widehat{u_t-\triangle u}=\widehat{f(x,t)} \]
we obtain an initial value problem for an ordinary differential equation:
dû 2
ˆ
+ |ξ | û = f (ξ, t)
dt
ˆ
û (ξ, 0) = ϕ (ξ).
The solution is given by

$$\widehat{u}(\xi,t)=e^{-|\xi|^2 t}\widehat{\phi}(\xi)+\int_0^t\ e^{-|\xi|^2(t-\tau)}\widehat{f}(\xi,\tau)\ d\tau.\]
Applying the inverse Fourier transform and a calculation as in the proof of Theorem 5.1, step (vi), we get}
2
−n/2 ix⋅ξ −|ξ| t ˆ
u(x, t) = (2π ) ∫ e (e ϕ (ξ)
n
R
t
2
−|ξ| (t−τ) ˆ(ξ, τ ) dτ ) dξ.
+ ∫ e f
0
From the above calculation for the homogeneous problem and calculation as in the proof of Theorem 5.1, step (vi), we obtain the
formula
1 2
−|y−x| /(4t)
u(x, t) = −
− ∫ ϕ(y)e dy
n
(2 √πt) R
n
t
1 2
−|y−x| /(4(t−τ))
+∫ ∫ f (y, τ ) e dy dτ .
−−− −−− − n
n
0 R (2 √ π(t − τ ) )
This function u(x, t) is a solution of the above inhomogeneous initial value problem provided
$$\phi\in C(\mathbb{R}^n),\ \ \sup_{\mathbb{R}^n}|\phi(x)|<\infty\]
and if
$$f\in C(\mathbb{R}^n\times[0,\infty)),\ \ M(\tau):=\sup_{\mathbb{R}^n}|f(y,\tau)|<\infty,\ 0\le\tau<\infty.\]



6.3: Maximum Principle
Let Ω ⊂ R be a bounded domain. Set
n
DT = Ω × (0, T ), T > 0,
ST = {(x, t) : (x, t) ∈ Ω × {0} or (x, t) ∈ ∂Ω × [0, T ]},
see Figure 6.3.1.
Figure 6.3.1: Notations to the maximum principle
Theorem 6.2
Assume u ∈ C (D ¯
¯¯¯¯¯
¯
T ) , that u , u t xi xk exist and are continuous in D , and T
$$u_t-\triangle u\le 0\ \ \mbox{in}\ D_T.\]

Then
$$\max_{\overline{D_T}} u(x,t)=\max_{S_T} u.\]
Proof
¯¯¯¯¯¯¯¯¯¯
¯ ¯¯¯¯¯¯¯¯¯¯
¯
Assume initially u t − △u < 0 in D . Let ε > 0 be small and
T 0 <ε <T . Since u ∈ C (DT −ε ) , there is an (x0 , t0 ) ∈ DT −ε
such that
$$u(x_0,t_0)=\max_{\overline{D_{T-\varepsilon}}} u(x,t).\]
Case (i)
Let (x 0, t0 ) ∈ DT −ε . Hence, since D T −ε is open,
ut (x0 , t0 ) = 0 ,u xl (x0 , t0 ) = 0 , l = 1, … , n and
$$\sum_{l,k=1}^n u_{x_lx_k}(x_0,t_0)\zeta_l\zeta_k\le0\ \ \mbox{for all}\ \zeta\in\mathbb{R}^n.\]
The previous inequality implies that u xk xk (x0 , t0 ) ≤ 0 for each k . Thus we arrived at a contradiction to u t − △u < 0 in D . T
Case (ii)
Assume (x , t ) ∈ Ω × {T − ε} . Then it follows as above △u ≤ 0 in (x , t ), and from
0 0 0 0 u(x0 , t0 ) ≥ u(x0 , t) , t ≤ t0 , one
concludes that u (x , t ) ≥ 0 . We arrived at a contradiction to u − △u < 0 in D again.
t 0 0 t T
Summarizing, we have shown that

$$\max_{\overline{D_{T-\varepsilon}}} u(x,t)=\max_{T-\varepsilon} u(x,t).\]
Thus there is an (x ε, tε ) ∈ ST −ε such that
$$u(x_\varepsilon,t_\varepsilon)=\max_{\overline{D_{T-\varepsilon}}} u(x,t).\]
Since u is continuous on D , we have ¯¯¯
¯
T
$$\lim_{\varepsilon\to 0}\max_{\overline{D_{T-\varepsilon}}} u(x,t)=\max_{\overline{D_T}} u(x,t).\]

It follows that there is (x, t) ∈ S such that
¯¯
¯ ¯
T
$$u(\overline{x},\overline{t})=\max_{\overline{D_T}} u(x,t)\]
since ST −ε ⊂ ST and S T is compact. Thus, theorem is shown under the assumption ut − △u < 0 in DT . Now assume
ut − △u ≤ 0 in D . Set T

$$v(x,t):=u(x,t)-kt,\]
where k is a positive constant. Then
$$v_t-\triangle v=u_t-\triangle u-k<0.\]
From above we have
max u(x, t) = max(v(x, t) + kt)

¯
¯¯¯
¯¯¯
¯ ¯
¯¯¯
¯¯¯
¯
DT DT
≤ max v(x, t) + kT
¯
¯¯¯
¯¯¯
¯
DT
= max v(x, t) + kT
ST
≤ max u(x, t) + kT ,
ST
Letting k → 0 , we obtain
$$\max_{\overline{D_T}} u(x,t)\le\max_{S_T} u(x,t).\]
¯
¯¯¯¯¯
¯
Since S T ⊂ DT , the theorem is shown.
□
If we replace in the above theorem the bounded domain Ω by R , then the result remains true provided we assume an {\it
n
additional} growth assumption for u. More precisely, we have the following result which is a corollary of the previous theorem. Set
for a fixed T , 0 < T < ∞ ,
$$D_T=\{ (x,t):\ x\in\mathbb{R}^n,\ 0<t<T\}.\]
Proposition 6.2
Assume u ∈ C (D ¯
¯¯¯¯¯
T
¯
) , that u , u
t xi xk exist and are continuous in D , T
$$u_t-\triangle u\le 0\ \ \mbox{in}\ D_T,\]

and additionally that u satisfies the growth condition
$$u(x,t)\le Me^{a|x|^2},\]
where M and a are positive constants.
Then
It follows immediately the
Corollary
The initial value problem u − △u = 0 in D , u(x, 0) = f (x), x ∈ R , has a unique solution in the class defined by
t T
n
u ∈ C (D ) , u , u exist and are continuous in D and |u(x, t)| ≤ M e .

¯
¯¯¯¯¯
¯ a|x|
T t xi xk T
Proof of Proposition 6.2.

See [10], pp. 217. We can assume that 4aT < 1 , since the finite interval can be divided into finite many intervals of equal
length τ with
4aτ < 1 . Then we conclude successively for k that
$$u(x,t)\le\sup_{y\in\mathbb{R}^n}u(y,k\tau)\le\sup_{y\in\mathbb{R}^n}u(y,0)\]
for kτ ≤ t ≤ (k + 1)τ , k = 0, … , N − 1 , where N = T /τ .
There is an ϵ > 0 such that 4a(T + ϵ) < 1 . Consider the comparison function

2
−n/2 |x−y | /(4(T +ϵ−t))
vμ (x, t) : = u(x, t) − μ (4π(T + ϵ − t)) e
= u(x, t) − μK(ix, iy, T + ϵ − t)
for fixed y ∈ R and for a constant μ > 0 . Since the heat kernel K(ix, iy, t) satisfies K
n
t = △Kx , we obtain
$$\frac{\partial}{\partial t}v_\mu-\triangle v_\mu=u_t-\triangle u\le0.\]
Set for a constant ρ > 0
$$D_{T,\rho}=\{(x,t):\ |x-y|<\rho,\ 0<t<T\}.\]
Then we obtain from Theorem 6.2 that
$$v_\mu(y,t)\le\max_{S_{T,\rho}}v_\mu,\]
where S T ,ρ ≡ ST of Theorem 6.2 with Ω = B ρ (y) , see Figure 6.3.1.
On the bottom of S T ,ρ we have, since μK > 0 ,
$$v_\mu(x,0)\le u(x,0)\le\sup_{z\in\mathbb{R}^n}f(z).\]
On the cylinder part |x − y| = ρ , 0 ≤ t ≤ T , of S T ,ρ it is
2
2
a|x| −n/2 ρ /(4(T +ϵ−t))
vμ (x, t) ≤ M e − μ (4π(T + ϵ − t)) e
2 2
a(|y|+ρ) −n/2 ρ /(4(T +ϵ))
≤ Me − μ (4π(T + ϵ)) e
≤ sup f (z)
n
z∈R
for all ρ > ρ (μ) , ρ sufficiently large. We recall that 4a(T + ϵ) < 1 .
0 0
Summarizing, we have
$$\max_{S_{T,\rho}}v_\mu(x,t)\le\sup_{z\in\mathbb{R}^n}f(z)\]
if ρ > ρ 0 (μ) . Thus
$$v_\mu(y,t)\le\max_{S_{T,\rho}}v_\mu(x,t)\le\sup_{z\in\mathbb{R}^n}f(z)\]
if ρ > ρ 0 (μ) .
Since
$$v_\mu(y,t)=u(y,t)-\mu\left(4\pi(T+\epsilon-t)\right)^{-n/2}\]
it follows
$$u(y,t)-\mu\left(4\pi(T+\epsilon-t)\right)^{-n/2}\le\sup_{z\in\mathbb{R}^n}f(z).\]
Letting μ → 0 , we obtain the assertion of the proposition.
□
The above maximum principle of Theorem 6.2 holds for a large class of parabolic differential operators, even for degenerate
equations.
Set
$$Lu=\sum_{i,j=1}^na^{ij}(x,t)u_{x_ix_j},\]
where a ij
∈ C (DT ) are real, a ij
=a
ji
, and the matrix (a ij
) is non-negative, that is,
$$\sum_{i,j=1}^na^{ij}(x,t)\zeta_i\zeta_j\ge 0\ \ \mbox{for all}\ \zeta\in\mathbb{R}^n,\]
and (x, t) ∈ D . T

Theorem 6.3
¯
¯¯¯¯¯
¯
Assume u ∈ C (D T ) , that u , u
t xi xk exist and are continuous in D , and T
$$u_t-L u\le 0\ \ \mbox{in}\ D_T.\]

Then
Proof
(i) One proof is a consequence of the following lemma: let A , B real, symmetric and non-negative matrices. Non-negative
n
means that all eigenvalues are non-negative. Then trace~(AB) ≡ ∑ a b ≥ 0 , see an exercise.
i,j=1
ij
ij
(ii) Another proof is more directly: let U = (z , … , z ) , where z is an orthonormal system of eigenvectors to the eigenvalues
1 n l
λ of the matrix A = (a (x , t )) . Set ζ = U η , x = U (x − x )y and v(y) = u(x + U y, t ) , then

i,j T
l 0 0 0 0 0
n n
ij 2
0 ≤ ∑ a (x0 , t0 )ζi ζj = ∑ λi η
i
i,j=1 i=1
n n
2
0 ≥ ∑ uxi xj ζi ζj = ∑ vy yi η .
i i
i,j=1 i=1
It follows λ
i ≥0 and v yi yi ≤0 for all i.
Consequently
$$\sum_{i,j=1}^na^{ij}(x_0,t_0)u_{x_ix_j}(x_0,t_0)=\sum_{i=1}^n\lambda_iv_{y_iy_i}\le0.\]
□



6.4: Initial-Boundary Value Problems
Consider the initial-boundary value problem for c = c(x, t)
ct = D△c in Ω × (0, ∞) (6.4.1)
¯
¯¯¯
c(x, 0) = c0 (x) x ∈ Ω (6.4.2)
∂c
= 0 on ∂Ω × (0, ∞). (6.4.3)
∂n
Here is Ω ⊂ R , n the exterior unit normal at the smooth parts of ∂Ω , D a positive constant and c
n
0 (x) a given function.
Remark. In application to diffusion problems, c(x, t) is the concentration of a substance in a solution, c0 (x) its initial
concentration and D the coefficient of diffusion.
The first Fick's rule says that
w = D∂c/∂n, (6.4.4)
where w is the flow of the substance through the boundary ∂Ω . Thus according to the Neumann boundary condition (6.4.3), we
assume that there is no flow through the boundary.



6.4.1: Fourier's 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 = λv in Ω (6.4.1.1)
∂v
= 0 on ∂Ω, (6.4.1.2)
∂n
and to the ordinary differential equation

′
w (t) + λDw(t) = 0. (6.4.1.3)
Assume Ω is bounded and ∂Ω sufficiently regular, then the eigenvalues of (6.4.1.1), (6.4.1.2) are countable and
$$0=\lambda_0<\lambda_1\le\lambda_2\le\ldots\to\infty.\]
Let v (x) be a complete system of orthonormal (in L
j
2
(Ω) ) eigenfunctions.
Solutions of (6.4.1.3) are
$$w_j(t)=C_je^{-D\lambda_jt},\]
where C are arbitrary constants.
j
According to the superposition principle,

$$c_N(x,t):=\sum_{j=0}^N C_je^{-D\lambda_jt}v_j(x)\]
is a solution of the differential equation (6.4.1.1) and
$$c(x,t):=\sum_{j=0}^\infty C_je^{-D\lambda_jt}v_j(x),\]
with
$$C_j=\int_\Omega\ c_0(x)v_j(x)\ dx,\]
is a formal solution of the initial-boundary value problem (6.4.1)-(6.4.3).
Diffusion in a tube
Consider a solution in a tube, see Figure 6.4.1.1.
Figure 6.4.1.1: Diffusion in a tube

Assume the initial concentration c (x , x , x ) of the substrate in a solution is constant if x = const. It follows from a
0 1 2 3 3
uniqueness result below that the solution of the initial-boundary value problem c(x , x , x , t) is independent of x and x .
1 2 3 1 2
Set z = x , then the above initial-boundary value problem reduces to

3
ct = Dczz
c(z, 0) = c0 (z)
cz = 0, z = 0, z = l.
The (formal) solution is

$$c(z,t)=\sum_{n=0}^\infty C_ne^{-D\left(\frac{\pi}{l}n\right)^2 t}\cos\left(\frac{\pi}{l}nz\right),\]

where
l
1
C0 = ∫ c0 (z) dz
l 0
l
2 π
Cn = ∫ c0 (z) cos( nz) dz, n ≥ 1.
l 0
l



6.4.2: Uniqueness
Sufficiently regular solutions of the initial-boundary value problem (6.4.1)-(6.4.3) are uniquely determined since from
ct = D△c in Ω × (0, ∞)
c(x, 0) = 0
∂c
= 0 on ∂Ω × (0, ∞).
∂n
it follows that for each τ >0
0 = ∫ ∫ (ct c − D(△c)c) dxdt

0 Ω
τ τ
1 ∂ 2 2
= ∫ ∫ (c ) dtdx + D ∫ ∫ | ∇x c | dxdt
Ω 0
2 ∂t Ω 0
τ
1 2
2
= ∫ c (x, τ ) dx + D ∫ ∫ | ∇x c | dxdt.
2 Ω Ω 0



6.5: Black-Scholes Equation
Solutions of the Black-Scholes equation define the value of a derivative, for example of a call or put option, which is based on an
asset. An asset can be a stock or a derivative of it, for instance. In principle, there are infinitely many such products, for example n-
th derivatives. The Black-Scholes equation for the value V (S, t) of a derivative is
1
2 2
Vt + σ S VSS + rS VS − rV = 0 in Ω, (6.5.1)
2
where for a fixed T , 0 < T <∞ ,

$$\Omega=\{(S,t)\in\mathbb{R}^2:\ 0<S<\infty,\ 0<t<T\}, \]
and σ, r are positive constants. More precisely, σ 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 Ito's Lemma under some assumptions on the random function associated to S(t) , see
[26], for instance.
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 (6.5.1):
C (S, T ) = max{S − E, 0} (6.5.2)
C (0, t) = 0 (6.5.3)
C (S, t) = S + o(S) as S → ∞, uniformly in t, (6.5.4)
where E and T are positive constants, E is the exercise price and T the expiry.
Side condition (6.5.2) means that the value of the option has no value at time T if S(T ) ≤ E , condition (6.5.3) says that it makes
no sense to buy assets if the value of the asset is zero, condition (6.5.4) means that we buy assets if its value becomes large, see
Figure 6.5.1, where the side conditions are indicated.
Figure 6.5.1: Side conditions for a call option

Theorem 6.4 (Black-Scholes formula for European call options).
The solution C (S, t), 0 ≤ S < ∞ , 0 ≤ t ≤ T , of the initial-boundary value problem (6.5.1)-(6.5.4) is explicitly known and is
given by
−r(T −t)
C (S, t) = SN (d1 ) − E e N (d2 ), (6.5.5)
where
x
1 −y
2
/2
N (x) = −− ∫ e dy,
√2π −∞
2
ln(S/E) + (r + σ /2)(T − t)
d1 = −−− − ,
σ√ T − t
2
ln(S/E) + (r − σ /2)(T − t)
d2 = .
−−− −
σ√ T − t

Proof. Substitutions
$$S=Ee^x,\ \ t=T-\frac{\tau}{\sigma^2/2},\ \ C=Ev(x,\tau)\]
change equation (6.5.1) to
vτ = vxx + (k − 1)vx − kv, (6.5.6)
where
$$k=\frac{r}{\sigma^2/2}.\]
Initial condition (6.5.6) implies
x
v(x, 0) = max{ e − 1, 0}. (6.5.7)
For a solution of (6.5.6) we make the ansatz

$$v=e^{\alpha x+\beta \tau}u(x,\tau),\]
where α and β are constants which we will determine as follows. Inserting the ansatz into differential equation (6.5.6), we get
$$\beta u+u_\tau=\alpha^2u+2\alpha u_x+u_{xx}+(k-1)(\alpha u+u_x)-ku.\]
Set β = α 2
+ (k − 1)α − k and choose α such that 0 = 2α + (k − 1) , then u τ = uxx . Thus
2
−(k−1)x/2−(k+1 ) τ/4
v(x, τ ) = e u(x, τ ), (6.5.8)
where u(x, τ ) is a solution of the initial value problem

uτ = uxx , − ∞ < x < ∞, τ > 0
u(x, 0) = u0 (x),
with
$$u_0(x)=\max\left\{e^{(k+1)x/2}-e^{(k-1)x/2},0\right\}.\]
A solution of this initial value problem is given by Poisson's formula
$$u(x,\tau)=\frac{1}{2\sqrt{\pi \tau}}\int_{-\infty}^{+\infty}\ u_0(s)e^{-(x-s)^2/(4\tau)}\ ds.\]
−−
Changing variable by q = (s − x)/(√2τ ) , we get
+∞
1 −− 2
−q /2
u(x, τ ) = −− ∫ u0 (q √2τ + x)e dq
√2π −∞
= I1 − I2 ,
where
∞
1 2
(k+1)(x+q√2τ ) −q /2
I1 = −− ∫ e e dq
√2π −x/( √2τ)
∞
1 2
(k−1)(x+q√2τ ) −q /2
I2 = −− ∫ e e dq.
√2π −x/( √2τ)
An elementary calculation shows that

2
(k+1)x/2+(k+1 ) τ/4
I1 = e N (d1 )
2
(k−1)x/2+(k−1 ) τ/4
I2 = e N (d2 ),
where

x 1 −−
d1 = −− + (k + 1)√2τ
√2τ 2
x 1 −−
d2 = −− + (k − 1)√2τ
√2τ 2
di
1 2
−s /2
N (di ) = −− ∫ e ds, i = 1, 2.
√2π −∞
Combining the formula for u(x, τ ), definition (6.5.8) of v(x, τ ) and the previous settings x = ln(S/E) , 2
τ = σ (T − t)/2 and
C = Ev(x, τ ) , we get finally the formula of Theorem 6.4.
In general, the solution u of the initial value problem for the heat equation is not uniquely defined, see for example [10], pp. 206.
Uniqueness. The uniqueness follows from the growth assumption (6.5.4). Assume there are two solutions of (6.5.1), (6.5.2)-(
6.5.4), then the difference W (S, t) satisfies the differential equation (6.5.1) and the side conditions
$$W(S,T)=0,\ W(0,t)=0,\ W(S,t)=O(S)\ \mbox{as}\ S\to\infty\]

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 of S and t . From the definition of u we see that
1
−αx−βτ
u(x, τ ) = e W (S, t), (6.5.9)
E
where S = Ee , t = T − 2τ /(σ
x 2
) . Thus we have the growth property
a|x| 1
|u(x, τ )| ≤ M e , x ∈ R , (6.5.10)
with positive constants M and a . Then the solution of u = u , in −∞ < x < ∞ ,

τ xx
0 ≤ τ ≤ σ T /2 , with the initial condition u(x, 0) = 0 is uniquely defined in the class of functions satisfying the growth condition
2
(6.5.10), see Proposition 6.2 of this chapter.

That is, u(x, τ ) ≡ 0 .
□
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 (6.5.1):
P (S, T ) = max{E − S, 0} (6.5.11)
−r(T −t)
P (0, t) = Ee (6.5.12)
P (S, t) = o(S) as S → ∞, uniformly in 0 ≤ t ≤ T . (6.5.13)
Here E is the exercise price and T the expiry.

Side condition (6.5.11) means that the value of the option has no value at time T if S(T ) ≥ E , condition (6.5.12) says that it
makes no sense to sell assets if the value of the asset is zero, condition (6.5.13) 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 solution P (S, t), 0 < S < ∞ , t < T of the initial-boundary value problem (6.5.1), (6.5.11)-(6.5.13) is explicitly known and
is given by
−r(T −t)
P (S, t) = E e N (−d2 ) − SN (−d1 ) (6.5.14)
where N (x), d , d are the same as in Theorem 6.4.

1 2

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(T-t)}\]
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.
□



6.E: Parabolic Equations (Exercises)
These are homework exercises to accompany Miersemann's "Partial Differential Equations" Textmap. This is a textbook targeted for a one
semester first course on differential equations, aimed at engineering students. Partial differential equations are differential equations that
contains unknown multivariable functions and their partial derivatives. Prerequisite for the course is the basic calculus sequence.
Q6.1
Show that the solution u(x, t) given by Poisson's formula satisfies
\[ \inf_{z\in \mathbb{R}^n}\varphi(z)\le u(x,t)\le\sup_{z\in\mathbb{R}^n}\varphi(z)\ , $$ provided φ(x) is continuous and bounded on
R .
n
Q6.2
Solve for given f (x) and μ ∈ R the initial value problem
1
1 1
ut + ux + μuxxx = 0 in R × R+
u(x, 0) = f (x) .
Q6.3
Show by using Poisson's formula: (i) Each function f ∈ C ([a, b]) can be approximated uniformly by a sequence f n ∈ C
∞
[a, b] . (ii) In (i)
we can choose polynomials f (Weierstrass's approximation theorem).
n
2 2
|y−x| |y−x|
Hint: Concerning (ii), replace the kernel K = exp(− 4t
) by a sequence of Taylor polynomials in the variable z = − 4t
.
Q6.4
Let u(x, t) be a positive solution of
′
ut = μuxx , t > 0, $$where$μ$isaconstant. Showthat$θ := −2μux /u$isasolutionof Burger sequation$$ θt (6.E.1)
+ θθx = μθxx , t > 0.
Q6.5
Assume u1 (s, t), . . . , un (s, t) are solutions of ut = uss . Show that ∏
n
k=1
uk (xk , t) is a solution of the heat equation ut − △u = 0 in
R
n
× (0, ∞) .
Q6.6
Let A , B are real, symmetric and non-negative matrices. Non-negative means that all eigenvalues are non-negative. Prove that trace
n
≥0 .
ij
(AB) ≡ ∑ a b ij
i,j=1
Hint: (i) Let U = (z1 , … , zn ) , where z is an orthonormal system of eigenvectors to the eigenvalues λ of the matrix B . Then
l l
−−
√λ1 0 ⋯ 0
⎛ ⎞
−−
⎜0 √λ2 ⋯ 0 ⎟
T
X =U ⎜ ⎟U (6.E.2)
⎜ ⎟
⎜⋯ ⋯ ⋯ ⋯ ⎟
−−
⎝ √λn ⎠
0 0 ⋯
is a square root of B . We recall that

λ1 0 ⋯ 0
⎛ ⎞
T ⎜0 λ2 ⋯ 0⎟
U BU = ⎜ ⎟. (6.E.3)
⎜⋯ ⋯ ⋯ ⋯⎟
⎝ ⎠
0 0 ⋯ λn
(ii) trace (QR) = trace (RQ). (iii) Let μ (C ), … μ (C ) are the eigenvalues of a real symmetric
1 n n×n -matrix. Then trace
C =∑
n
l=1
μl (C ) , which follows from the fundamental lemma of algebra:
n n−1
det (λI − C ) = λ − (c11 + … + cnn )λ +…
≡ (λ − μ1 ) ⋅ … ⋅ (λ − μn )
n n+1
= λ − (μ1 + … + μn )λ +…

Q6.7
Assume Ω is bounded, u is a solution of the heat equation and u satisfies the regularity assumptions of the maximum principle (Theorem
6.2). Show that u achieves its maximum and its minimum on S . T
Q6.8
Prove the following comparison principle: Assume Ω is bounded and u, v satisfy the regularity assumptions of the maximum principle.
Then
ut − △u ≤ vt − △v in DT
u ≤ v on ST
imply that u ≤ v in D . T
Q6.9
Show that the comparison principle implies the maximum principle.
Q6.10
Consider the boundary-initial value problem
ut − △u = f (x, t) in DT
u(x, t) = ϕ(x, t) on ST ,
¯
¯¯¯¯¯
¯
where f , ϕ are given.\\ Prove uniqueness in the class u, u t, ux
i
xk ∈ C (DT ) .
Q6.11
¯
¯¯¯¯¯
¯
Assume u, v 1, v2 ∈ C
2
(DT ) ∩ C (DT ) , and u is a solution of the previous boundary-initial value problem and v , v satisfy 1 2
(v1 )t − △v1 ≤ f (x, t) ≤ (v2 )t − △v2 in DT
v1 ≤ ϕ ≤ v2 on ST .
Show that (inclusion theorem)

¯
¯¯¯¯¯
¯
v1 (x, t) ≤ u(x, t) ≤ v2 (x, t) on DT . (6.E.4)
Q6.12
Show by using the comparison principle: let u be a sufficiently regular solution of
ut − △u = 1 in DT
u = 0 on ST ,
then 0 ≤ u(x, t) ≤ t in DT .
Q6.13
Discuss the result of Theorem 6.3 for the case
n n
Lu = ∑ aij (x, t)uxi xj + ∑ bi (x, t)uxi + c(x, t)u(x, t). (6.E.5)
i,j=1 i
Q6.14
Show that
∞
2
−n t
u(x, t) = ∑ cn e sin(nx), (6.E.6)
n=1
where
π
2
cn = ∫ f (x) sin(nx) dx, (6.E.7)
π 0
is a solution of the initial-boundary value problem

ut = uxx , x ∈ (0, π), t > 0,
u(x, 0) = f (x),
u(0, t) = 0,
u(π, t) = 0,
if f ∈ C
4
(R) is odd with respect to 0 and 2π-periodic.
Q6.15
(i) Find the solution of the diffusion problem c = Dc in 0 ≤ z ≤ l$, $0 ≤ t < ∞ ,
t zz D = const. > 0 , under the boundary conditions
c (z, t) = 0 if z = 0 and z = l and with the given initial concentration
z
c0 = const. 0 ≤z ≤h
c(z, 0) = c0 (z) := { (6.E.8)
0 h < z ≤ l.
(ii) Calculate lim t→∞ c(z, t) .
Q6.16
Solve the initial-boundary value problem (rotationally symmetric solution in a ball): find c(r, t) on (0, R) × (0, ∞) of
∂c 1 ∂ 2
∂c
= (Dr ) − kc (6.E.9)
2
∂t r ∂r ∂r
c(r, 0) = h(r), 0 < r < R, (6.E.10)
c(R, t) = c0 (boundary condition), (6.E.11)
sup |c(r, t)| < ∞ (boundary condition), (6.E.12)

0<r<R, 0<t<T
where T >0 is fixed, k , c , D are positive constants, and

0
⎧ 0 0 < r < R0
h(r) = ⎨ r−R0 , (6.E.13)
⎩ c0 R0 < r < R
R−R0
where 0 < R 0 <R and R close to R .

0
Q6.17
Prove the Black-Scholes formula for an European put option.
Hint: Put-call parity.
Q6.18
Prove the put-call parity for European options
−r(T −t)
C (S, t) − P (S, t) = S − Ee (6.E.14)
by using the following uniqueness result: Assume W is a solution of (6.5.1) 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 .
Q6.19
Prove that a solution V (S, t) of the initial-boundary value problem (6.5.1) in Ω 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
(Ω) ∩ C (Ω) .
Q6.20
Prove that a solution V (S, t) of the initial-boundary value problem (6.5.1) in Ω, 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 , satisfies |V (S, t)| ≤ cS for all S ≥ 0 and
0 ≤t ≤T .



This page is licensed under a not declared license and was authored, remixed, and/or curated by Erich Miersemann. A detailed versioning
history of the edits to source content is available upon request.

CHAPTER OVERVIEW
7: ELLIPTIC EQUATIONS OF SECOND ORDER
Here we consider linear elliptic equations of second order, mainly the Laplace equation
$$ \triangle 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 ∈ Ω ⊂ R , is n
$$Lu:=\sum_{i,j=1}^na^{ij}(x)u_{x_ix_j}+\sum_{j=1}^n b^j(x)u_{x_j}+c(x)u=f(x),\]
where the matrix A = (a ij

) is real, symmetric and positive definite. If A is a constant matrix, then a transform to principal axis and
stretching of axis leads to
$$\sum_{i,j=1}^na^{ij}u_{x_ix_j}=\triangle v,\]
where v(y) ,
:= u(T y) T stands for the above composition of mappings.

7.1: FUNDAMENTAL SOLUTION

7.2: REPRESENTATION FORMULA
7.2.1: CONCLUSIONS FROM THE REPRESENTATION FORMULA
7.3.1: BOUNDARY VALUE PROBLEMS: DIRICHLET PROBLEM
7.3.2: BOUNDARY VALUE PROBLEMS: NEUMANN PROBLEM
7.3.3: BOUNDARY VALUE PROBLEMS: MIXED BOUNDARY VALUE PROBLEM
7.4: GREEN'S FUNCTION FOR Δ
7.4.1: GREEN'S FUNCTION FOR A BALL
7.4.2: GREEN'S FUNCTION AND CONFORMAL MAPPING
7.5: INHOMOGENEOUS EQUATION
7.E: ELLIPTIC EQUATIONS OF SECOND ORDER (EXERCISES)
1 3/30/2022
7.1: Fundamental Solution
Here we consider particular solutions of the Laplace equation in R of the type n
$$u(x)=f(|x-y|),\]
where y ∈ R is fixed and f is a function which we will determine such that u defines a solution if the Laplace equation.
n
Set r = |x − y| , then
xi − yi
′
ux = f (r)
i
r
2 2
(xi − yi ) 1 (xi − yi )
′′ ′
ux xi = f (r) + f (r) ( − )
i
2 3
r r r
′′
n−1 ′
△u = f (r) + f (r).
r
Thus a solution of △u = 0 is given by

$$f(r)=\left\{
c1 ln r + c2 n =2
(7.1.1)
2−n
c1 r + c2 n ≥3
\right.\]
with constants c , c .
1 2
Definition. Set r = |x − y| . The function

$$
s(r):=\left\{
1
− ln r n =2
2π
2−n (7.1.2)
r
n ≥3
(n−2)ωn
\right.
\]
is called singularity function associated to the Laplace equation. Here is ωn the area of the n-dimensional unit sphere which is
given byω = 2π /Γ(n/2), where
n
n/2
$$\Gamma(t):=\int_0^\infty\ e^{-\rho}\rho^{t-1}\ d\rho,\ \ t>0,\]

is the Gamma function.
Definition. A function
$$\gamma(x,y)=s(r)+\phi(x,y)\]
is called fundamental solution associated to the Laplace equation if ϕ ∈ C 2
(Ω) and △ xϕ =0 for each fixed y ∈ Ω .
Remark. The fundamental solution γ satisfies for each fixed y ∈ Ω the relation
$$-\int_\Omega\ \gamma(x,y)\triangle_x\Phi(x)\ dx=\Phi(y)\ \ \mbox{for all}\ \Phi\in C_0^2(\Omega),\]
see an exercise. This formula follows from considerations similar to the next section.
In the language of distribution, this relation can be written by definition as
$$-\triangle_x\gamma(x,y)=\delta(x-y),\]
where δ is the Dirac distribution, which is called δ -function.



7.2: Representation Formula
In the following we assume that Ω, the function ϕ which appears in the definition of the fundamental solution and the potential
function u considered are sufficiently regular such that the following calculations make sense, see [6] for generalizations. This is
¯
¯¯¯ ¯
¯¯¯
the case if Ω is bounded, ∂Ω is in C , ϕ ∈ C (Ω) for each fixed y ∈ Ω and u ∈ C (Ω) .
1 2 2
Figure 7.2.1: Notations to Green's identity

Theorem 7.1. Let u be a potential function and γ a fundamental solution, then for each fixed y ∈ Ω
∂u(x) ∂γ(x, y)
u(y) = ∫ (γ(x, y) − u(x) ) dSx . (7.2.1)
∂Ω
∂nx ∂nx
Proof. Let B ρ (y) ⊂Ω be a ball. Set Ω ρ (y) = Ω ∖ Bρ (y) . See Figure 7.2.2 for notations.
Figure 7.2.2: Notations to Theorem 7.1

From Green's formula, for u, v ∈ C 2 ¯
¯¯
(Ω)
¯
,
∂u ∂v
∫ (v△u − u△v) dx = ∫ (v −u ) dS (7.2.2)
Ωρ (y) ∂ Ωρ (y) ∂n ∂n
we obtain, if v is a fundamental solution and u a potential function,

∂u ∂v
∫ (v −u ) dS = 0. (7.2.3)
∂ Ωρ (y) ∂n ∂n
Thus we have to consider

∂u ∂u ∂u
∫ v dS = ∫ v dS + ∫ v dS
∂ Ωρ (y) ∂n ∂Ω ∂n ∂ Bρ (y) ∂n
∂v ∂v ∂v
∫ u dS = ∫ u dS + ∫ u dS.
∂ Ωρ (y)
∂n ∂Ω
∂n ∂ Bρ (y)
∂n
We estimate the integrals over ∂ B ρ (y) :

(i)
∣ ∂u ∣
∣∫ v dS ∣ ≤ M ∫ |v| dS
∣ ∂ Bρ (y)
∂n ∣ ∂ Bρ (y)
n−1
≤ M (∫ s(ρ) dS + C ωn ρ ),
∂ Bρ (y)

where
M = M (y) = sup |∂u/∂n|, ρ ≤ ρ0 ,
Bρ (y)
0
C = C (y) = sup |ϕ(x, y)|.

x∈Bρ (y)
0
From the definition of s(ρ) we get the estimate as ρ → 0
∂u O(ρ| ln ρ|) n =2
∫ v dS = { (7.2.4)
∂ Bρ (y) ∂n O(ρ) n ≥ 3.
(ii) Consider the case n ≥ 3 , then

∂v 1 1 ∂ϕ
∫ u dS = ∫ u dS + ∫ u dS
n−1
∂ Bρ (y)
∂n ωn ∂ Bρ (y)
ρ ∂ Bρ (y)
∂n
1
n−1
= ∫ u dS + O(ρ )
n−1
ωn ρ ∂ Bρ (y)
1
n−1
= u(x0 ) ∫ dS + O(ρ ),
n−1
ωn ρ ∂ Bρ (y)
n−1
= u(x0 ) + O(ρ ).
for an x0 ∈ ∂ Bρ (y) .
Combining this estimate and (7.2.4), we obtain the representation formula of the theorem.
□
Corollary. Set ϕ ≡ 0 and r = |x − y| in the representation formula of Theorem 7.1, then

1 ∂u ∂(ln r)
u(y) = ∫ (ln r −u ) dSx , n = 2, (7.2.5)
2π ∂Ω ∂nx ∂nx
2−n
1 1 ∂u ∂(r )
u(y) = ∫ ( −u ) dSx , n ≥ 3. (7.2.6)
n−2
(n − 2)ωn ∂Ω r ∂nx ∂nx



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 (7.2.5), (7.2.6). We recall that a function u is called harmonic if u ∈ C (Ω) and 2
△u = 0 in Ω .
Proposition 7.1. Assume u is harmonic in Ω. Then u ∈ C ∞

(Ω) .}
Proof. Let Ω ⊂⊂ Ω be a domain such that y ∈ Ω . It follows from representation formulas (7.2.5), (7.2.6), where Ω := Ω , that
0 0 0
D u(y) exist and are continuous for all l since one can change differentiation with integration in right hand sides of the
l
representation formulas.
□
Remark. In fact, a function which is harmonic in Ω is even real analytic in Ω, see an exercise.
Proposition 7.2 (Mean value formula for harmonic functions). Assume u is harmonic in Ω. Then for each B ρ (x) ⊂⊂ Ω
1
u(x) = ∫ u(y) dSy . (7.2.1.1)
n−1
ωn ρ ∂ Bρ (x)
Proof. Consider the case n ≥ 3 . The assertion follows from (7.2.6) where Ω := B ρ (x) since r = ρ and
1 ∂u 1 ∂u
∫ dSy = ∫ dSy
n−2 n−2
∂ Bρ (x) r ∂ny ρ ∂ Bρ (x)
∂ny
1
= ∫ △u dy
ρn−2 Bρ (x)
= 0.
We recall that a domain Ω ∈ R is called connected if Ω is not the union of two nonempty open subsets
n
Ω1 , Ω2 such that
Ω ∩ Ω = ∅ . A domain in R is connected if and only if its path connected.
n
1 2
Proposition 7.3 (Maximum principle). Assume u is harmonic in a connected domain and achieves its supremum or infimum in Ω.
Then u ≡ const. in Ω.
Proof. Consider the case of the supremum. Let x 0 ∈ Ω such that
u(x0 ) = sup u(x) =: M . (7.2.1.2)
Ω
Set
Ω1 := {x ∈ Ω : u(x) = M } and Ω 2 := {x ∈ Ω : u(x) < M } . The set Ω is not empty since x
1 0 ∈ Ω1 . The set Ω is open since
2
¯¯¯¯¯¯¯¯¯¯¯¯¯
¯
u ∈ C
2
. Consequently, Ω is empty if we can show that Ω is open. Let x ∈ Ω , then there is a ρ > 0 such that
(Ω) 2 1
¯¯
¯
1 0 Bρ (x)
0
¯¯
¯
⊂Ω
and u(x) = M for all x ∈ B (x̄) . If not, then there exists ρ > 0 and x̂ such that
ρ
0
¯
¯
| x̂ − x̄| = ρ , 0 < ρ < ρ

¯
¯
0 and u(x̂) < M . From the mean value formula, see Proposition 7.2, it follows
1 M
M = ∫ u(x) dS < ∫ dS = M , (7.2.1.3)
n−1 n−1
ωn ρ ¯¯
∂ Bρ ( x̄) ωn ρ ¯¯
∂ Bρ ( x̄)
which is a contradiction. Thus, the set Ω is empty since Ω is open.

2 1
¯
¯¯¯
Corollary. Assume Ω is connected and bounded, and u ∈ C 2
(Ω) ∩ C (Ω) is harmonic in Ω. Then u achieves its minimum and its
maximum on the boundary ∂Ω .
Remark. The previous corollary fails if Ω is not bounded as simple counterexamples show.



7.3.1: Boundary Value Problems: Dirichlet Problem
¯
¯¯¯
The Dirichlet problem (first boundary value problem) is to find a solution u ∈ C 2
(Ω) ∩ C (Ω) of
△u = 0 in Ω (7.3.1.1)
u = Φ on ∂Ω, (7.3.1.2)
where Φ is given and continuous on ∂Ω .

Proposition 7.4. Assume Ω 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 condition (7.3.1.2) one point from the the boundary as the
following example shows. Let Ω ⊂ R be the domain
2
$$
\Omega=\{x\in B_1(0):\ x_2>0\},
\]
Figure 7.3.1.1: Counterexample

¯
¯¯¯
Assume u ∈ C 2
(Ω) ∩ C (Ω ∖ {0}) is a solution of
△u = 0 in Ω
u = 0 on ∂Ω ∖ {0}.
This problem has solutions u ≡ 0 and u = Im(z + z −1

) , where z = x 1 + i x2 . Concerning another example see an exercise.
In contrast to this behavior of the Laplace equation, one has uniqueness if $\triangle u=0$ is replaced by the minimal surface
equation
$$
\frac{\partial}{\partial x_1}\left(\frac{u_{x_1}}{\sqrt{1+|\nabla u|^2}}\right)+
\frac{\partial}{\partial x_2}\left(\frac{u_{x_2}}{\sqrt{1+|\nabla u|^2}}\right)=0.
\]



7.3.2: Boundary Value Problems: Neumann Problem
¯
¯¯¯
The Neumann problem (second boundary value problem) is to find a solution u ∈ C 2
(Ω) ∩ C
1
(Ω) of
△u = 0 in Ω (7.3.2.1)
∂u
= Φ on ∂Ω, (7.3.2.2)
∂n
where Φ is given and continuous on ∂Ω .

¯
¯¯¯
Proposition 7.5. Assume Ω is bounded, then a solution to the Dirichlet problem is in the class u ∈ C 2
(Ω) uniquely determined up
to a constant.
Proof. Exercise. Hint: Multiply the differential equation △w = 0 by w and integrate the result over Ω.
¯
¯¯¯
Another proof under the weaker assumption u ∈ C (Ω) ∩ C
1 2
(Ω) follows from the Hopf boundary point lemma, see Lecture
Notes: Linear Elliptic Equations of Second Order, for instance.



7.3.3: Boundary Value Problems: Mixed Boundary Value Problem
¯
¯¯¯
The Mixed boundary value problem (third boundary value problem) is to find a solution u ∈ C 2
(Ω) ∩ C
1
(Ω) of
△u = 0 in Ω (7.3.3.1)
∂u
+ hu = Φ on ∂Ω, (7.3.3.2)
∂n
where Φ and h are given and continuous on ∂Ω .e Φ and h are given and continuous on ∂Ω .
Proposition 7.6. Assume Ω is bounded and sufficiently regular, then a solution to the mixed problem is uniquely determined in the
¯
¯¯¯
class u ∈ C 2
(Ω) provided h(x) ≥ 0 on ∂Ω and h(x) > 0 for at least one point x ∈ ∂Ω .
Proof. Exercise. Hint: Multiply the differential equation △w = 0 by w and integrate the result over Ω.



7.4: Green's Function for Δ Δ
Theorem 7.1 says that each harmonic function satisfies
∂u(y) ∂γ(y, x)
u(x) = ∫ (γ(y, x) − u(y) ) dSy , (7.4.1)
∂Ω
∂ny ∂ny
where γ(y, x) is a fundamental solution. In general, u does not satisfies the boundary condition in the above boundary value
problems. Since γ = s + ϕ , see Section 7.2, where ϕ is an arbitrary harmonic function for each fixed x, we try to find a ϕ such
that u satisfies also the boundary condition.
Consider the Dirichlet problem, then we look for a ϕ such that
γ(y, x) = 0, y ∈ ∂Ω, x ∈ Ω. (7.4.2)
Then
∂γ(y, x)
u(x) = − ∫ u(y) dSy , x ∈ Ω. (7.4.3)
∂Ω
∂ny
Suppose that u achieves its boundary values Φ of the Dirichlet problem, then
∂γ(y, x)
u(x) = − ∫ Φ(y) dSy , (7.4.4)
∂Ω
∂ny
We claim that this function solves the Dirichlet problem (7.3.1.1), (7.3.1.2).
A function γ(y, x) which satisfies (7.4.2), and some additional assumptions, is called Green's function. More precisely, we define a
Green function as follows.
¯
¯¯¯
Definition. A function G(y, x), y, x ∈ Ω , x ≠ y , is called Green function associated to Ω and to the Dirichlet problem (7.3.1.1),
(7.3.1.2) if for fixed x ∈ Ω, that is we consider G(y, x) as a function of y , the following properties hold:
¯
¯¯¯
(i) G(y, x) ∈ C 2
(Ω ∖ {x}) ∩ C (Ω ∖ {x}) ,△ y G(y, x) = 0, x ≠ y .
¯
¯¯¯
(ii) G(y, x) − s(|x − y|) ∈ C 2
(Ω) ∩ C (Ω) .
(iii) G(y, x) = 0 if y ∈ ∂Ω , 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 definition two further important properties, provided Ω is bounded,
sufficiently regular and connected.
Proposition 7.7. A Green function has the following properties. In the case n = 2 we assume {\rm diam} Ω < 1 .
(A) G(x, y) = G(y, x)\ \ (symmetry).
(B) 0 < G(x, y) < s(|x − y|), x, y ∈ Ω, x ≠ y .
¯
¯¯¯
¯¯
Proof. (A) Let x
(1) (2)
, x ∈ Ω . Set Bi = Bρ (x
(i)
) , i = 1, 2 . We assume Bi ⊂ Ω and B1 ∩ B2 = ∅ . Since G(y, x
(1)
) and
¯
¯¯¯¯
¯ ¯
¯¯¯¯
¯
G(y, x
(2)
) are harmonic in Ω ∖ (B 1 ∪ B2 ) we obtain from Green's identity, see Figure 7.4.1 for notations,

Figure 7.4.1: Proof of Proposition 7.7
(1)
∂ (2)
0 = ∫ (G(y, x ) G(y, x )
¯
¯¯¯
¯¯¯
∂(Ω∖( B1 ∪B2 ))
¯
¯¯¯
¯¯¯
∂ny
∂
(2) (1)
−G(y, x ) G(y, x ))dSy
∂ny
∂ ∂
(1) (2) (2) (1)
= ∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy
∂Ω ∂ny ∂ny
(1)
∂ (2) (2)
∂ (1)
+ ∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy
∂B1
∂ny ∂ny
∂ ∂
(1) (2) (2) (1)
+ ∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy .
∂B2 ∂ny ∂ny
The integral over ∂Ω is zero because of property (iii) of a Green function, and
(1)
∂ (2) (2)
∂ (1)
∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x ))dSy
∂B1 ∂ny ∂ny
(1) (2)
→ G(x ,x ),
∂ ∂
(1) (2) (2) (1)
∫ (G(y, x ) G(y, x ) − G(y, x ) G(y, x )) dSy
∂B2
∂ny ∂ny
(2) (1)
→ −G(x ,x )
as ρ → 0 .
This follows by considerations as in the proof of Theorem 7.1.
(B) Since
G(y, x) = s(|x − y|) + ϕ(y, x) (7.4.5)
and G(y, x) = 0 if y ∈ ∂Ω and x ∈ Ω we have for y ∈ ∂Ω
ϕ(y, x) = −s(|x − y|). (7.4.6)
From the definition of s(|x − y|) it follows that ϕ(y, x) < 0 if y ∈ ∂Ω . Thus, since △y ϕ = 0 in Ω , the maximum-minimum
principle implies that ϕ(y, x) < 0
for all y, x ∈ Ω. Consequently
G(y, x) < s(|x − y|), x, y ∈ Ω, x ≠ y. (7.4.7)
It remains to show that

G(y, x) > 0, x, y ∈ Ω, x ≠ y. (7.4.8)
Fix x ∈ Ω and let B ρ (x) be a ball such that B ρ (x) ⊂Ω for all 0 < ρ < ρ . There is a sufficiently small ρ
0 0 >0 such that for each
ρ, 0 < ρ < ρ ,
0

¯
¯¯¯¯¯¯¯¯¯¯¯
¯
G(y, x) > 0 for all y ∈ Bρ (x), x ≠ y, (7.4.9)
see property (iii) of a Green function. Since

¯
¯¯¯¯¯¯¯¯¯¯¯
¯
△y G(y, x) = 0 in Ω ∖ Bρ (x)
G(y, x) > 0 if y ∈ ∂ Bρ (x)
G(y, x) = 0 if y ∈ ∂Ω
it follows from the maximum-minimum principle that

$$
G(y,x)>0\ \ \mbox{on}\ \Omega\setminus\overline{B_\rho(x)}.
\]
□



7.4.1: Green's Function for a Ball
If Ω = B R (0) is a ball, then Green's function is explicitly known.
Let Ω = B (0) be a ball in R with radius R and the center at the origin. Let x, y ∈ B
R
n
R (0) and let y the reflected point of y on
′
the sphere ∂ B (0) , that is, in particular |y||y | = R , see Figure 7.4.1.1 for notations.
R
′ 2
Figure 7.4.1.1: Reflection on ∂ B R (0)
Set
$$G(x,y)=s(r)-s\left(\frac{\rho}{R}r_1\right),\]
where s is the singularity function of Section 7.1, r = |x − y| and
$$\rho^2=\sum_{i=1}^ny_i^2,\ \ \ r_1=\sum_{i=1}^n\left(x_i-\frac{R^2}{\rho^2}y_i\right)^2.\]
This function G(x, y) satisfies (i)-(iii) of the definition of a Green function. We claim that
$$u(x)=-\int_{\partial B_R(0)} \frac{\partial}{\partial n_y}G(x,y)\Phi\ dS_y\]
is a solution of the Dirichlet problem (7.3.1.1), (7.3.1.2). This formula is also true for a large class of domains Ω ⊂ R , see [13]. n
Lemma.
$$-\frac{\partial}{\partial n_y}G(x,y)\bigg|_{|y|=R}=\frac{1}{R\omega_n}\frac{R^2-|x|^2}{|y-x|^n}.\]
Proof. Exercise.
Set
2 2
1 R − |x |
H (x, y) = n
, (7.4.1.1)
Rωn |y − x|
which is called Poisson's kernel.

Theorem 7.2. Assume Φ ∈ C (∂Ω) . Then
u(x) = ∫ H (x, y)Φ(y) dSy (7.4.1.1)

∂ BR (0)
¯
¯¯¯
is the solution of the first boundary value problem (7.3.1.1), (7.3.1.2) in the class C 2
(Ω) ∩ C (Ω) .
Proof. The proof follows from following properties of H (x, y):
i. H (x, y) ∈ C , |y| = R, |x| < R, x ≠ y ,
∞
ii. △ H (x, y) = 0, |x| < R, |y| = R ,

x
iii. ∫ H (x, y) dS = 1, |x| < R ,

∂ BR (0)
y
iv. H (x, y) > 0, |y| = R, |x| < R ,

v. Fix ζ ∈ ∂ B (0) and δ > 0 , then lim
R x→ζ,|x|<R H (x, y) = 0 uniformly in y ∈ ∂ B R (0), |y − ζ| > δ .
(i), (iv) and (v) follow from the definition (7.4.1.1) of H and (ii) from (7.4.1.1) or from
$$H=-\frac{\partial G(x,y)}{\partial n_y}\bigg|_{y\in\partial B_R(0)},\]
G harmonic and G(x, y) = G(y, x).
Property (iii) is a consequence of formula
$$u(x)=\int_{\partial 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 Poisson's formula, is in C (BR (0)) and that u achieves the prescribed boundary values. Fix
ζ ∈ ∂ B (0) and let x ∈ B (0). Then
R R
u(x) − Φ(ζ) = ∫ H (x, y) (Φ(y) − Φ(ζ)) dSy

∂ BR (0)
= I1 + I2 ,
where
I1 = ∫ H (x, y) (Φ(y) − Φ(ζ)) dSy

∂ BR (0), |y−ζ|<δ
I2 = ∫ H (x, y) (Φ(y) − Φ(ζ)) dSy .

∂ BR (0), |y−ζ|≥δ
For given (small) ϵ > 0 there is a δ = δ(ϵ) > 0 such that

$$|\Phi(y)-\Phi(\zeta)|<\epsilon\]
for all y ∈ ∂ B R (0) with |y − ζ| < δ . It follows |I 1| ≤ϵ because of (iii) and (iv). Set M = max∂ BR (0) |ϕ| . From (v) we conclude
that there is a δ ′
> 0 such that
$$H(x,y)<\frac{\epsilon}{2M\omega_nR^{n-1}}\]
if x and y satisfy |x − ζ| < δ ′
, |y − ζ| > δ , see Figure 7.4.1.2 for notations.
Figure 7.4.1.2: Proof of Theorem 7.2

Thus |I 2| <ϵ and the inequality
$$|u(x)-\Phi(\zeta)|<2\epsilon\]
for x ∈ B R (0) such that |x − ζ| < δ is shown.
′
Remark. Define δ ∈ [0, π] through cos δ = x ⋅ y/(|x||y|), then we write Poisson's formula of Theorem 7.2 as
$$u(x)=\frac{R^2-|x|^2}{\omega_nR}\int_{\partial B_R(0)}\Phi(y)\frac{1}{\left(|x|^2+R^2-2|x|R\cos\delta\right)^{n/2}}\ dS_y.\]
In the case n = 2 we can expand this integral in a power series with respect to ρ := |x|/R if |x| < R , since
2 2 2
R − |x | 1 −ρ
=
2 2
|x| + R − 2|x|R cos δ ρ − 2ρ cos δ + 1
∞
n
= 1 +2 ∑ρ cos(nδ),
n=1
see [16], pp. 18 for an easy proof of this formula, or [4], Vol. II, p. 246.



7.4.2: Green's Function and Conformal Mapping
For two-dimensional domains there is a beautiful connection between conformal mapping and Green's function. Let w = f (z) be a
conformal mapping from a sufficiently regular connected domain in R onto the interior of the unit circle, see Figure 7.4.2.1
2
Figure 7.4.2.1: Conformal mapping

Then the Green function of Ω is, see for example [16] or other text books about the theory of functions of one complex variable,
¯
¯¯¯¯¯¯¯¯¯
¯
∣ ∣
1 1 − f (z)f (z0 )
∣ ∣
G(z, z0 ) = ln , (7.4.2.1)
∣ ∣
2π f (z) − f (z0 )
∣ ∣
where z = x 1 + i x2 ,z
0 = y1 + i y2 .



7.5: Inhomogeneous Equation
¯
¯¯¯
Here we consider solutions u ∈ C 2
(Ω) ∩ C (Ω) of
−△u = f (x) in Ω (7.5.1)
u = 0 on ∂Ω, (7.5.2)
where f is given.
We need the following lemma concerning volume potentials. We assume that Ω is bounded and sufficiently regular such that all the
following integrals exist. See [6] for generalizations concerning these assumptions.
Let for x ∈ R , n ≥ 3 ,
n
1
V (x) = ∫ f (y) dy (7.5.3)
n−2
Ω |x − y|
and set in the two-dimensional case

1
V (x) = ∫ f (y) ln( ) dy. (7.5.4)
Ω
|x − y|
We recall that ω n = |∂ B1 (0)| .

Lemma.
(i) Assume f ∈ C (Ω) . Then V ∈ C
1 n
(R ) and
∂ 1
Vxi (x) = ∫ f (y) ( ) dy, if n ≥ 3,
n−2
Ω
∂xi |x − y|
∂ 1
Vxi (x) = ∫ f (y) (ln( )) dy if n = 2.
Ω
∂xi |x − y|
(ii) If f ∈ C
1
(Ω) , then V ∈ C
2
(Ω) and
△V = −(n − 2)ωn f (x), x ∈ Ω, n ≥ 3
△V = −2πf (x), x ∈ Ω, n = 2.
Proof. To simplify the presentation, we consider the case n = 3 .

(i) The first assertion follows since we can change differentiation with integration since the differentiate integrand is weakly
singular, see an exercise.
(ii) We will differentiate at x ∈ Ω. Let B be a fixed ball such that x ∈ B , ρ sufficiently small such that B
ρ ρ ρ ⊂Ω . Then, according
to (i)
and since we have the identity
∂ 1 ∂ 1
( ) =− ( ) (7.5.5)
∂xi |x − y| ∂yi |x − y|
which implies that

∂ 1 ∂ 1 1
f (y) ( ) =− (f (y) ) + fy (y) , (7.5.6)
i
∂xi |x − y| ∂yi |x − y| |x − y|
we obtain

∂ 1
Vxi (x) = ∫ f (y) ( ) dy
Ω
∂xi |x − y|
∂ 1 ∂ 1
= ∫ f (y) ( ) dy + ∫ f (y) ( ) dy
Ω∖Bρ
∂xi |x − y| Bρ
∂xi |x − y|
∂ 1
= ∫ f (y) ( ) dy
Ω∖Bρ ∂xi |x − y|
∂ 1 1
+∫ (− (f (y) ) + fy (y) ) dy
i
Bρ ∂yi |x − y| |x − y|
∂ 1
= ∫ f (y) ( ) dy
Ω∖Bρ ∂xi |x − y|
1 1
+∫ fy (y) dy − ∫ f (y) ni dSy ,
i
Bρ |x − y| ∂Bρ |x − y|
where n is the exterior unit normal at ∂B . It follows that the first and second integral is in C
ρ
1
. The second integral is also in
(Ω)
C (Ω) according to (i) and since f ∈ C (Ω) by assumption.

1 1
−1
Because of △ x (|x − y| ) = 0, x ≠ y , it follows
n
∂ 1
△V = ∫ ∑ fy (y) ( ) dy
i
Bρ
∂xi |x − y|
i=1
n
∂ 1
−∫ f (y) ∑ ( ) ni dSy .
∂Bρ
∂xi |x − y|
i=1
Now we choose for B a ball with the center at x, then

ρ
△V = I1 + I2 , (7.5.7)
where
n
yi − xi
I1 = ∫ ∑ fy (y) dy
i
3
Bρ (x) i=1 |x − y|
1
I2 = − ∫ f (y) dSy .
2
∂ Bρ (x) ρ
We recall that n ⋅ (y − x) = ρ if y ∈ ∂ B ρ (x) . It is I 1 = O(ρ) as ρ → 0 and for I we obtain from the mean value theorem of the
2
integral calculus that for a ȳ ∈ ∂ B (x)

¯
¯
ρ
1
¯¯
I2 = − f (ȳ )∫ dSy
2
ρ ∂ Bρ (x)
¯¯
= −ωn f (ȳ ),
which implies that lim ρ→0 I2 = −ωn f (x) .

□
In the following we assume that Green's function exists for the domain Ω, which is the case if Ω is a ball.
¯
¯¯¯
Theorem 7.3. Assume f ∈ C
1
(Ω) ∩ C (Ω) . Then
u(x) = ∫ G(x, y)f (y) dy (7.5.8)

Ω

is the solution of the inhomogeneous problem (7.5.1), (7.5.2).
Proof. For simplicity of the presentation let n = 3 . We will show that
u(x) := ∫ G(x, y)f (y) dy (7.5.9)

Ω
is a solution of (7.3.1.1), (7.3.1.2). Since

1
G(x, y) = + ϕ(x, y), (7.5.10)
4π|x − y|
where ϕ is a potential function with respect to x or y , we obtain from the above lemma that
1 1
△u = △∫ f (y) dy + ∫ △x ϕ(x, y)f (y) dy
4π Ω |x − y| Ω
= −f (x),
where x ∈ Ω. It remains to show that u achieves its boundary values. That is, for fixed x 0 ∈ ∂Ω we will prove that
lim u(x) = 0. (7.5.11)
x→x0 , x∈Ω
Set
u(x) = I1 + I2 , (7.5.12)
where
I1 (x) = ∫ G(x, y)f (y) dy,

Ω∖ Bρ ( x0 )
I2 (x) = ∫ G(x, y)f (y) dy.

Ω∩Bρ ( x0 )
Let M = max¯¯¯¯ |f (x)|

Ω
. Since
1 1
G(x, y) = + ϕ(x, y), (7.5.13)
4π |x − y|
we obtain, if x ∈ B ρ (x0 ) ∩Ω ,
M dy
2
| I2 | ≤ ∫ + O(ρ )
4π Ω∩Bρ ( x0 ) |x − y|
M dy
2
≤ ∫ + O(ρ )
4π B2ρ( x) |x − y|
2
= O(ρ )
as ρ → 0 . Consequently for given ϵ there is a ρ 0 = ρ0 (ϵ) > 0 such that

ϵ
| I2 | < for all 0 < ρ ≤ ρ0 . (7.5.14)
2
For each fixed ρ, 0 < ρ ≤ ρ , we have

0

lim I1 (x) = 0 (7.5.15)
x→x0 , x∈Ω
since G(x 0, y) = 0 if y ∈ Ω ∖ B
ρ (x0 ) and G(x, y) is uniformly continuous in x ∈ B ρ/2
(x0 ) ∩ Ω and y ∈ Ω ∖ B ρ (x0 ) , see Figure
7.5.1.
Figure 7.5.1: Proof of Theorem 7.3

□
Remark. For the proof of (ii) in the above lemma it is sufficient to assume that f is H\"older continuous. More precisely, let
f ∈ C (Ω) , 0 < λ < 1 , then V ∈ C (Ω) , see for instance [9].
λ 2,λ



7.E: Elliptic Equations of Second Order (Exercises)
calculus sequence.
Q7.1
Let γ(x, y) be a fundamental solution to △, y ∈ Ω . Show that
$$
-\int_\Omega\gamma(x,y)\ \triangle\Phi(x)\ dx=\Phi(y)\quad\hbox{for all}\ \
\Phi\in C_0^2(\Omega)\ .
\]
Hint: See the proof of the representation formula.
Q7.2
Show that |x | sin(k|x|) is a solution of the Helmholtz equation
−1
$$
\triangle u+k^2u=0 \ \mbox{in}\ \mathbb{R}^n\setminus\{0\}.
\]
Q7.3
¯
¯¯¯
Assume u ∈ C 2
(Ω) , Ω bounded and sufficiently regular, is a solution of
3
△u = u in Ω
u = 0 on ∂Ω.
Show that u = 0 in Ω.
Q7.4
Let Ω α = {x ∈ R
2
: x1 > 0, 0 < x2 < x1 tan α} , 0 < α ≤ π . Show that
π
k
π
u(x) = r α
sin( kθ) (7.E.1)
α
is a harmonic function in Ω satisfying u = 0 on ∂Ω , provided k is an integer. Here (r, θ) are polar coordinates with the center at
α α
(0, 0).
Q7.5
¯
¯¯¯
Let u ∈ C
2
(Ω) be a solution of △u = 0 on the quadrangle Ω = (0, 1) × (0, 1) satisfying the boundary conditions
¯
¯¯¯
u(0, y) = u(1, y) = 0 for all y ∈ [0, 1] and u (x, 0) = u (x, 1) = 0 for all x ∈ [0, 1]. Prove that u ≡ 0 in Ω.
y y
Q7.6
Let u ∈ C 2
(R )
n
be a solution of △u = 0 in R satisfying u ∈ L
n 2
(R )
n
, i. e., ∫ R
n
2
u (x) dx < ∞. Show that u ≡ 0 in R . n
Hint: Prove
2
const. 2
∫ |∇u | dx ≤ ∫ |u | dx, (7.E.2)
2
BR (0) R B2R (0)
where c is a constant independent of R .

To show this inequality, multiply the differential equation by ζ := η u , where 2
η ∈ C is a cut-off function with properties: η ≡ 1 in B (0), η ≡ 0 in the exterior of B

1
R 2R (0),
0 ≤ η ≤ 1 , |∇η| ≤ C /R. Integrate the product, apply integration by parts and use
the formula 2ab ≤ ϵa + b , ϵ > 0 . 2 1
ϵ
2
Q7.7
Show that a bounded harmonic function defined on R must be a constant (a theorem of Liouville).
n
Q7.8
¯¯¯¯¯¯¯¯¯¯¯
¯
Assume u ∈ C 2
(B1 (0)) ∩ C (B1 (0) ∖ {(1, 0)}) is a solution of
△u = 0 in B1 (0)
u = 0 on ∂ B1 (0) ∖ {(1, 0)}.
Show that there are at least two solutions.

Hint: Consider
$$
u(x,y)=\frac{1-(x^2+y^2)}{(1-x)^2+y^2}.
\]
Q7.9
Assume Ω ⊂R
n
is bounded and u, v ∈ C
2
(Ω) ∩ C (Ω)
¯
¯¯¯
satisfy △u = △v and max∂Ω |u − v| ≤ ϵ for given ϵ>0 . Show that
max¯¯¯¯ |u − v| ≤ ϵ .
Ω
Q7.10
¯¯¯¯¯¯¯¯¯¯¯
¯ ¯
¯¯¯
Set Ω = R ∖ B (0) and let u ∈ C (Ω) be a harmonic function in Ω satisfying lim
n
1
2
|x|→∞ u(x) = 0 . Prove that
$$
\max_{\overline{\Omega}}|u|=\max_{\partial\Omega}|u|\ .
\]
Hint: Apply the maximum principle to Ω ∩ B R (0) , R large.
Q7.11
Let Ωα = {x ∈ R
2
: x1 > 0, 0 < x2 < x1 tan α} , 0 <α ≤π , Ωα,R = Ωα ∩ BR (0) , and assume f is given and bounded on
¯
¯¯¯¯¯¯¯¯
¯
Ωα,R .
¯
¯¯¯¯¯¯¯¯
¯
Show that for each solution u ∈ C (Ω ) ∩ C (Ω 1
α,R
2
α,R ) of △u = f in
Ω satisfying u = 0 on ∂ Ω ∩ B (0) , holds:
α,R α,R R
For given ϵ > 0 there is a constant C (ϵ) such that

$$
|u(x)|\le C(\epsilon)\ |x|^{{\pi\over\alpha}-\epsilon}\qquad\hbox{in}\
\Omega_{\alpha,R}.
\]
¯
¯¯¯ ¯
¯¯¯
Hint: (a) Comparison principle (a consequence from the maximum principle): Assume Ω is bounded, u, v ∈ C
2
(Ω) ∩ C (Ω)
satisfying −△u ≤ −△v in Ω and u ≤ v on ∂Ω . Then u ≤ v in Ω.

(b) An appropriate comparison function is
π
−ϵ
v = Ar α
sin(B(θ + η)) , (7.E.3)
A, B, η appropriate constants, B, η positive.

Q7.12
¯
¯¯¯
Let Ω be the quadrangle (−1, 1) × (−1, 1) and u ∈ C (Ω) ∩ C (Ω) a solution of the boundary value problem
2
−△u = 1 in Ω ,
u =0 on ∂Ω . Find a lower and an upper bound for u(0, 0).
Hint: Consider the comparison function v = A(x 2
+y )
2
, A = const.
Q7.13
¯¯¯¯¯¯¯¯¯¯¯
¯
Let u ∈ C (B (0)) ∩ C (B (0)) satisfying u ≥ 0, △u = 0 in B
2
a a a (0) . Prove (Harnack's inequality):
$$
{a^{n-2}(a-|\zeta|)\over (a+|\zeta|)^{n-1}}u(0)\le u(\zeta)\le
{a^{n-2}(a+|\zeta|)\over (a-|\zeta|)^{n-1}}u(0)\ .
\]
Hint: Use the formula (see Theorem 7.2)
2 2
a − |y | u(x)
u(y) = ∫ dSx (7.E.4)
n
aωn |x|=a |x − y|
for y = ζ and y = 0 .
Q7.14
Let ϕ(θ) be a 2π-periodic C -function with the Fourier series
4
ϕ(θ) = ∑ (an cos(nθ) + bn sin(nθ)) . (7.E.5)
n=0
Show that
∞
n
u = ∑ (an cos(nθ) + bn sin(nθ)) r (7.E.6)
n=0
solves the Dirichlet problem in B 1 (0) .
Q7.15
Assume u ∈ C 2
(Ω) satisfies △u = 0 in Ω. Let B a (ζ) be a ball such that its closure is in Ω.
Show that
|α|
|α|γn
α
|D u(ζ)| ≤ M ( ) , (7.E.7)
a
where M = sup x∈B

a (ζ)
|u(x)| and γ n = 2nωn−1 /((n − 1)ωn ) .
Hint: Use the formula of Theorem 7.2, successively to the k th derivatives in balls with radius a(|α| − k)/m , k = o, 1, … , m − 1 .
Q7.16
Use the result of the previous exercise to show that u ∈ C 2
(Ω) satisfying △u = 0 in Ω is real analytic in Ω.
Hint: Use Stirling's formula
n −n −−− 1
n! = n e ( √2πn + O ( − )) (7.E.8)
√n

as n → ∞ , to show that u is in the class C (ζ), where K = cM and r = a/(eγ ) . The constant c is the constant in the estimate
K,r n
n ≤ c e n! which follows from Stirling's formula. See Section 3.5 for the definition of a real analytic function.
n n
Q7.17
Assume Ω is connected and u ∈ C (Ω) is a solution of △u = 0 in Ω. Prove that u ≡ 0 in Ω if D
2 α
u(ζ) = 0 for all α , for a point
ζ ∈ Ω . In particular, u ≡ 0 in Ω if u ≡ 0 in an open subset of Ω .
Q7.18
Let Ω = {(x 1, x2 , x3 ) ∈ R
3
: x3 > 0} , which is a half-space of R . Show that 3
1 1
G(x, y) = − , (7.E.9)
¯
¯¯
4π|x − y| 4π|x − y |
where ȳ = (y
¯
¯
1, y2 , −y3 ) , is the Green function to Ω.
Q7.19
Let Ω = {(x 1, x2 , x3 ) ∈ R
3
: x
2
1
+x
2
2
+x
2
3
2
< R , x3 > 0} , which is half of a ball in R . Show that 3
1 R
G(x, y) = −
4π|x − y| 4π|y||x − y ⋆ |
1 R
− + ,
¯
¯¯ ⋆
¯
¯¯
4π|x − y | 4π|y||x − y |
⋆
where ȳ = (y , y , −y ) , y
¯
¯
1 2 3
⋆ 2
= R y/(|y | )
2
and ȳ
¯
¯ ¯¯
= R ȳ
2
/(|y | )
2
,
is the Green function to Ω.
Q7.20
Let Ω = {(x 1, x2 , x3 ) ∈ R
3
: x2 > 0, x3 > 0} , which is a wedge in R . Show that 3
1 1
G(x, y) = −
¯¯
4π|x − y| 4π|x − ȳ |
1 1
− + ,
′ ′
4π|x − y | ¯¯
4π|x − ȳ |
′
where y = (y , y , −y ) , y
¯
¯¯
1 2 3
′
= (y1 , −y2 , y3 ) and y ¯
¯¯
= (y1 , −y2 , −y3 ) ,
is the Green function to Ω.
Q7.21
Find Green's function for the exterior of a disk, i. e., of the domain Ω = {x ∈ R 2
: |x| > R} .
Q7.22
Find Green's function for the angle domain Ω = {z ∈ C : 0 < arg z < απ} , 0 < α < π .
Q7.23
Find Green's function for the slit domain Ω = {z ∈ C : 0 < arg z < 2π} .
Q7.24
Let for a sufficiently regular domain Ω ∈ R , a ball or a quadrangle for example,
n

F (x) = ∫ K(x, y) dy, (7.E.10)
Ω
¯
¯¯¯ ¯
¯¯¯
where K(x, y) is continuous in Ω × Ω where x ≠ y , and which satisfies
c
|K(x, y)| ≤ (7.E.11)
α
|x − y|
with a constants c and α , α < n .

¯
¯¯¯
Show that F (x) is continuous on Ω.
Q7.25
Prove (i) of the lemma of Section 7.5.
Hint: Consider the case n ≥ 3 . Fix a function η ∈ C 1
(R )
1
satisfying 0 ≤η ≤1 , 0 ≤η
′
≤2 , η(t) = 0 for t ≤1 , η(t) = 1 for
t ≥ 2 and consider for ϵ > 0 the regularized integral
dy
Vϵ (x) := ∫ f (y)ηϵ , (7.E.12)
n−2
Ω |x − y|
where η = η(|x − y|/ϵ) . Show that V converges uniformly to

ϵ ϵ V on compact subsets of R
n
as ϵ→ 0 , and that ∂ Vϵ (x)/∂ xi
converges uniformly on compact subsets of R to n
∂ 1
∫ f (y) ( ) dy (7.E.13)
n−2
Ω ∂xi |x − y|
as ϵ → 0 .



Bibliography
[1] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical tables. Vol.
55, National Bureau of Standards Applied Mathematics Series, U.S. Government Printing Office,Washington, DC, 1964. Reprinted
by Dover, New York, 1972.
[2] S. Bernstein, Sur un théorème de géométrie et son application aux dérivées partielles du type elliptique. Comm. Soc. Math.
de Kharkov (2)15,(1915–1917), 38–45. German translation: Math. Z. 26 (1927), 551–558.
[3] E. Bombieri, E. De Giorgi and E. Giusti, Minimal cones and the Bernstein problem. Inv. Math. 7 (1969), 243–268.
[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 Differential Equations. Graduate Studies in Mathematics, 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 Differential Equations. Chelsia Publishing Company, New York, 1986.
[9] D. Gilbarg and N. S. Trudinger, Elliptic Partial Differential Equations of Second Order. Grundlehren, Vol. 224, Springer-Verlag,
Berlin, 1983. 201
[10] F. John, Partial Differential Equations. Springer-Verlag, New York,1982.
[11] K. Königsberger, 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 Russian. English translation: Course of Theoretical Physics. Vol. 1, Pergamon Press, Oxford, 1976.
[13] R. Leis, Vorlesungen über partielle Differentialgleichungen zweiter Ordnung. B. I.-Hochschultaschenbücher 165/165a,
Mannheim, 1967.
[14] J.-L. Lions and E. Magenes, Problémes aux limites non homogénes et applications. Dunod, Paris, 1968.
[15] E. Miersemann, Kapillarflächen. Ber. Verh. Sächs. 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 über Partielle Differentialgleichungen. Teubner, Leipzig, 1955. Translation from Russian.
Englisch translation: Lectures on Partial Differential 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), 62–105.
[20] W. I. Smirnow, Lehrgang der Höheren 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 Höheren Mathematik., Teil IV. VEB Verlag der Wiss., Berlin, 1975. Translation from Russian.
English translation: Course of Higher Mathematics, Vol. 4., Elsevier, 1964.
[22] A. Sommerfeld, Partielle Differentialgleichungen. Geest & Portig, Leipzig, 1954.
[23] W. A. Strauss, Partial Differential equations. An Introduction. Second edition, Wiley-Interscience, 2008. German translation:
Partielle Differentialgleichungen. Vieweg, 1995.
[24] M. E. Taylor, Pseudodifferential 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
B F L
C G P
Poisson's formula
DIRICHLET INTEGRAL
Glossary
Sample Word 1 | Sample Definition 1

Partial Differential Equations - LibreTexts

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

Partial Differential Equations - LibreTexts

Uploaded by

Copyright:

Available Formats

PARTIAL DIFFERENTIAL

This text was compiled on 03/30/2022

2: EQUATIONS OF FIRST ORDER

2.0: PRELUDE TO FIRST ORDER EQUATIONS

2.5: HAMILTON-JACOBI THEORY

4.1: ONE-DIMENSIONAL WAVE EQUATION

6.1: POISSON'S FORMULA

7: ELLIPTIC EQUATIONS OF SECOND ORDER

Picard-Lindelöf Theorem. Suppose

for all (x, y1 ), (x, y2 ) .

Figure 1.0.1: Initial value problem

= 0 in R has infinitely many linearly independent solutions in the linear space C (R ) .

The ordinary differential equation of second order

Erich Miersemann 1.0.1 3/30/2022 https://math.libretexts.org/@go/page/2193

where y and y are given, or to consider the boundary value problem

Figure 1.0.2: Boundary value problem

Contributors and Attributions

Contributions and Attributions

Erich Miersemann 1.0.2 3/30/2022 https://math.libretexts.org/@go/page/2193

∫ M (x, y)dx + N (x, y)dy (1.1.3)

Figure 1.1.1: Independence of the path

Example 1.1.4: Method of an integrating multiplier for an ordinary

Erich Miersemann 1.1.1 3/30/2022 https://math.libretexts.org/@go/page/2128

M (x, y)dx + N (x, y)dy = 0 (1.1.5)

u = H (v(x, y)), (1.1.7)

where H is an arbitrary C -function.

Example 1.1.6: Cauchy-Riemann equations

Example 1.1.7: Newton Potential

is a solution of the Laplace equation in R 3

Erich Miersemann 1.1.2 3/30/2022 https://math.libretexts.org/@go/page/2128

where △u = u x1 x1 + ux2 x2 + ux3 x3 and k is a positive constant. The condition

where h(x, t) is given, is a boundary condition for the heat equation.

Example 1.1.9: Wave equation

Figure 1.1.2: Oscillating string

describes oscillations of a string.

Contributors and Attributions

Erich Miersemann 1.1.3 3/30/2022 https://math.libretexts.org/@go/page/2128

Contributions and Attributions

Erich Miersemann 1.1.4 3/30/2022 https://math.libretexts.org/@go/page/2128

$$V=\{v\in C^2[a,b]:\ v(a)=u_a,\ v(b)=u_b\},\]

Exercise 1.2.1: Proof

Figure 1.2.1.1: Admissible Variations

Basic lemma in the calculus of variations. Let h ∈ C (a, b) and

Erich Miersemann 1.2.1 3/30/2022 https://math.libretexts.org/@go/page/2130

Contributors and Attributions

Contributions and Attributions

Erich Miersemann 1.2.2 3/30/2022 https://math.libretexts.org/@go/page/2130

Example 1.2.2.1: Dirichlet integral

Example 1.2.2.2: Minimal surface equation

Erich Miersemann 1.3.1 3/30/2022 https://math.libretexts.org/@go/page/2131

⎜ −−−−−−−−⎟+ ⎜ −−−−−−−− ⎟ = 0. (1.3.2)

Figure 1.2.2.2: Rotationally symmetric minimal surface

Neumann type boundary value problems

Assume u is a minimizer of E(v) in V , that is

Erich Miersemann 1.3.2 3/30/2022 https://math.libretexts.org/@go/page/2131

∫ ( ∑ Fu (x, u, ∇u)ϕx + Fu (x, u, ∇u)ϕ) dx

− ∫ gu (x, u)ϕ ds = 0

where ν = (ν , … , ν ) is the exterior unit normal at the boundary

Example 1.2.2.3: Laplace equation

Example 1.2.2.4: Capillary equation

Let Ω ⊂ R and set

$$E(v)=\int_\Omega\ \sqrt{1+|\nabla v|^2}\ dx+\frac{\kappa}{2}\int_\Omega\ v^2\ dx -\cos\gamma\int_{\partial\Omega}\ v\

Then the related boundary value problem is

where we use the abbreviation