Linear Elliptic Equations of Second Order: Lecture Notes

Linear Elliptic Equations of Second Order
Lecture Notes
Erich Miersemann
Department of Mathematics
Leipzig University
Version October, 2015

2
Contents
1 Potential theory 7
1.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Dipole potential . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Single layer potential . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Integral equations . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5 Volume potential . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2 Perron’s method 47
2.1 A maximum principle . . . . . . . . . . . . . . . . . . . . . . 47
2.2 Subharmonic, superharmonic functions . . . . . . . . . . . . . 50
2.3 Boundary behavior . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.1 Examples for local barriers . . . . . . . . . . . . . . . 59
2.4 Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3 Maximum principles 67
3.1 Basic maximum principles . . . . . . . . . . . . . . . . . . . . 67
3.1.1 Directional derivative boundary value problem . . . . 72
3.1.2 Behavior near a corner . . . . . . . . . . . . . . . . . . 73
3.1.3 An a priori estimate . . . . . . . . . . . . . . . . . . . 76
3.2 A discrete maximum principle . . . . . . . . . . . . . . . . . . 77
3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3
4 CONTENTS
Preface
These lecture notes present an introduction to linear second order elliptic

partial differential equations. It can be considered as a continuation of a
chapter on elliptic equations of the lecture notes [17] on partial differential
equations. In [17] we focused our attention mainly on explicit solutions for
standard problems for elliptic, parabolic and hyperbolic equations.
The first chapter concerns integral equation methods for boundary value
problems of the Laplace equation. This method can be extended to a large
class of linear elliptic equations and systems. In the following chapter we
consider Perron’s method for the Dirichlet problem for the Laplace equation.
This method is based on the maximum principle and on an estimates of
derivatives of solutions of the Laplace equation.
For additional reading we recommend following books: W. I. Smirnov [21],
I. G. Petrowski [20], D. Gilbarg and N. S. Trudinger [10], S. G. Michlin [14],
P. R. Garabedian [9], W. A. Strauss [22], F. John [13], L. C. Evans [5] and
R. Courant and D. Hilbert [4]. Some material of these lecture notes was
taken from some of these books.
5
6 CONTENTS
Chapter 1
Potential theory
The notation potential has its origin in Newton’s attraction rule

Mm y − x
K(x, y) = −G ,
|y − x|2 |y − x|
where G = 6.67 · 10−11 m3 /(kg · s2 ), and K is the force acting between two
mass points M and m located at x, y ∈ R3 , respectively. Since rot K = 0,
there is a scalar function Q(x, y), called potential, such that ∇x Q(x, y) =
K(x, y). Thus, Q(x, y) = −GM m|y − x|−1 is a Newton potential. The
function Q(x, y) defines the work which has to be done to move one of the
mass points to infinity if the other one is fixed.
Let Ω ⊂ Rn be a bounded, connected and sufficiently regular domain.
Consider for given f and h the boundary value problem
−△v = f in Ω
v = h on ∂Ω.
We can transform this problem into a boundary value problem for the
Laplace equation by setting v = u − w, where
Z
w(x) = s(|x − y|)f (y) dy.
Ω
Here s(r) denotes the singularity function, see also [17],

1
− 2π ln r : n=2

s(r) := r 2−n
(n−2)ωn
: n≥3
We recall that ωn = |∂B1 (0)|. Since w ∈ C 2 (Ω) and −△w = f in Ω if f is

sufficiently regular, see Section 7.5 in [17], we arrive at the problem △u = 0
7
8 CHAPTER 1. POTENTIAL THEORY
in Ω and v = h − w on ∂Ω. Consequently, it is sufficient to consider the

boundary value problem for the Laplace equation, which is a problem with
a homogeneous differential equation.
The Dirichlet problem (first boundary value problem) is to find a solution

u ∈ C 2 (Ω) ∩ C(Ω) of
△u = 0 in Ω (1.1)
u = Φ on ∂Ω, (1.2)
where Φ is given and continuous on ∂Ω.

The Neumann problem (second boundary value problem) is to find a
solution u ∈ C 2 (Ω) ∩ C 1 (Ω) of
△u = 0 in Ω (1.3)
∂u
= Ψ on ∂Ω, (1.4)
∂n
where Ψ is given and continuous on ∂Ω.
In [17], Chapter 7, we derived an explicit formula for the solution of (1.1),
(1.2) if Ω is a ball. In general, one gets explicit solutions, provided the Green
function is known for the domain Ω considered.
We denote (1.1), (1.2) by (Di ) and (1.3), (1.4) by (Ni ) to indicate that
the problems considered concerns the interior of Ω. Then (De ) and (Ne )
denote the associated exterior problems, that is, we have to replace in (1.1)
and (1.3) the domain Ω by its complement Rn \ Ω.
For the Dirichlet problems we make an ansatz with a dipole potential

∂ 1
Z
W (z) = σ(y) dSy (1.5)
∂Ω ∂ν(y) |z − y|n−2
if n ≥ 3. In the case that n = 2 we have to replace |z −y|2−n by − ln(|z −y|).

In the formula above ν(y) denotes the exterior unit normal at y ∈ ∂Ω and
σ(y) the dipole density.
For the Neumann problem we make an ansatz with a single layer potential
σ(y)
Z
V (z) = dSy (1.6)
∂Ω |z − y|n−2
if n ≥ 3. In the case that n = 2 we have to replace |z −y|2−n by − ln(|z −y|).

1.1. PRELIMINARIES 9
Both potentials solve the Laplace equation in Rn \ ∂Ω.
In the rest of this chapter we assume that n ≥ 3.

We will see that discontinuous properties of these surface potentials lead to
integral equations which can be studied by using Fredholm’s results on inte-
gral equations. Thus, the method of surface potentials provide a beautiful
example for Fredholm’s theory.
1.1 Preliminaries
Let Ω ⊂ Rn be a bounded and connected domain with a sufficiently regular
boundary ∂Ω.
Definition. We say that ∂Ω ∈ C 1,λ , 0 < λ ≤ 1, if:
(i) For each given x ∈ ∂Ω there exists a ρ > 0 and N = N (x, ρ) balls
B2ρ (xi ) ⊂ Rn , i = 1, . . . , N , with centers xi ∈ ∂Ω, where x1 = x, such that
N
[
∂Ω ⊂ Bρ (xi ).
i=1
(ii) Let Txi be a plane which contains xi and denote by Z2ρ (xi ) a circular
cylinder parallel to the normal on Txi such that its intersection with the
plane Txi is a ball in Rn−1 with radius 2ρ and the center at xi . We assume
that the intersection ∂Ω∩Z2ρ (xi ) has a local representation τ = f (ξ), f ≡ fi ,
where ξ is in an (n − 1)-dimensional ball D2ρ = D2ρ (0) with radius 2ρ and
the center at 0 ∈ Rn−1 . Moreover, we assume
f ∈ C 1,λ (D2ρ ), f (0) = 0, ∇f (0) = 0.
Lemma 1.1.1 (Partition of unity). There exists ηi ∈ C0∞ (B2ρ (xi )), 0 ≤
ηi ≤ 1, such that
N
X N
[
ηi (x) = 1 if x ∈ Bρ (xi ).
i=1 i=1
Proof. For given B2ρ (xi ) there exists φi ∈ C0∞ (B2ρ (xi )) with the properties
B2ρ(xi )
2ρ
Tx
i
x
i
δΩ
Z2ρ (xi )
Figure 1.1: Definition of ∂Ω ∈ C 1,λ
that φi = 1 in Bρ (xi ) and 0 ≤ φi (x) ≤ 1, see an exercise. Set
η1 = φ1
= 1 − (1 − φ1 ),
ηi = φi (1 − φ1 ) · . . . · (1 − φi−1 )
= (1 − (1 − φi ))(1 − φ1 ) · . . . · (1 − φi−1 ).
Then
N
X
ηi (x) = 1 − (1 − φ1 ) · . . . · (1 − φN ),
i=1
which implies that

N
X N
[
ηi (x) = 1, if x ∈ Bρ (xi )
i=1 i=1
since at least one of the factors is zero. ✷
Assume ∂Ω ∈ C 1,λ , then we define the area integral by

Z Z N
X
g(y) dSy = ηi (y) g(y) dSy
∂Ω ∂Ω i=1
N Z
X
= ηi (y) g(y) dSy ,
i=1 ∂Ω
1.2. DIPOLE POTENTIAL 11
where
Z
ηi (y) g(y) dSy
∂Ω
Z p
= ηi (ξ, fi (ξ)) g(ξ, fi (ξ)) 1 + |∇fi (ξ)|2 dξ.
D2ρ
Here we suppose that ∂Ω ∩ Z2ρ (xi ) has the parametric representation y =

(ξ, fi (ξ)).
1.2 Dipole potential

Dipole density. The following consideration leads to the formula (1.5) for
the dipole potential in the case of three dimensions.
Consider two parallel surfaces, one inside of Ω and the other one outside,
of distance ǫ/2, ǫ > 0 small, from ∂Ω, see Figure 1.2. Assume there is a
y+
ε/2
ν(y)
ε/2 y
y−
x δΩ
Figure 1.2: Double layer potential
charge of power ǫ−1 at y + = y + (ǫ/2)ν(y) and a charge of power −ǫ−1 at

y − = y − (ǫ/2)ν(y). Set z + = y + − x and z − = y − − x, then the potential
at x of the dipole is

1 1 1
u = −
ǫ |z + | |z − |
1 |z − |2 − |z + |2
= .
ǫ (|z + | + |z − |)|z + ||z − |
Since z − = z + − ǫν(y), we have
|z − |2 = hz + − ǫν(y), z + − ǫν(y)i
= |z + |2 − 2ǫz + · ν(y) + ǫ2 .
Thus
1 2ǫz + · ν(y) + ǫ2
u=− .
ǫ (|z + | + |z − |)|z + ||z − |
Set z = y − x, then
z · ν(y)
lim u = −
ǫ→0 |z|3

∂ 1
=
∂ν(y) |y − x|
is the potential of a single dipole with density σ(y) = 1 at y. Multiplication
with a density σ(y) and integration over ∂Ω leads to (1.5).
The right hand side of (1.5) is called dipole potential or potential of a double
layer with density σ. The dipole potential is in C ∞ (Rn \ ∂Ω) and a solution
of the Laplace equation in Rn \ ∂Ω. In fact, see the following proposition,
the right hand side of (1.5) is defined and continuous on ∂Ω provided the
boundary ∂Ω is sufficiently smooth, but W (x) makes a jump across ∂Ω.
Some of the following calculations are based on the formula for the di-
rectional derivative in direction ν(y)
n
∂ 1 n−2 X
= (zi − yi )(ν(y))i . (1.7)
∂ν(y) |z − y|n−2 |z − y|n
i=1
Lemma 1.2.1. Assume ∂Ω ∈ C 1,λ and σ ∈ C(∂Ω). Then the right hand
side of (1.5) is defined and is continuous on ∂Ω.
Proof. Consider the case n ≥ 3. Let x be the center of a local coordinate

system and z ∈ ∂Ω ∩ Z2ρ (x), see Figure 1.3. We have to show that, see
Section 1.1 for the definition of the surface integral and formula (1.7),
−(ζ − ξ) · ∇f (ξ) + f (ζ) − f (ξ)
Z
q(ζ) := η(ξ, f (ξ)) σ(ξ, f (ξ)) dξ
D2ρ (|ζ − ξ|2 + |f (ζ) − f (ξ)|2 )n/2
is continuous in a neighborhood of 0 ∈ Rn−1 . Here is D2ρ = D2ρ (0), z =

(ζ, f (ζ)) and y = (ξ, f (ξ)) in local coordinates, and η ∈ C ∞ in its arguments.
ν (y)
Tx
x
z y
δΩ
Z2ρ (x )
Figure 1.3: Local coordinates
Because of f ∈ C 1,λ (D2ρ ), f (0) = 0 and ∇f (0) = 0 we have
A(ξ, ζ)
Z
q(ζ) = n−1−λ
dξ,
D2ρ |ξ − ζ|
where A(ξ, ζ) is bounded on D2ρ × D2ρ and continuous if ξ 6= ζ. Since the

integrand is weakly singular, it follows that q(ζ) is continuous on D2ρ , see
an exercise. ✷
Let x0 ∈ ∂Ω and x ∈ Rn . Set
W (x) = W1 (x) + σ(x0 )W0 (x),
where

∂ 1
Z
W1 (x) = (σ(y) − σ(x0 )) dSy , (1.8)
∂Ω ∂ν(y) |x − y|n−2

∂ 1
Z
W0 (x) = dSy . (1.9)
∂Ω ∂ν(y) |x − y|n−2
The integral W0 (x) is called Gauss integral.
Lemma 1.2.2. Suppose that ∂Ω ∈ C 1,λ . Then


 −(n − 2)ωn : x∈Ω
W0 (x) = 0 : x 6∈ Ω .
n−2
− 2 ωn : x ∈ ∂Ω

Proof. (i) If x ∈ Rn \ Ω is fixed, then there is a domain Ω0 ⊃⊃ Ω where

|x − y|2−n ∈ C ∞ (Ω0 ) and satisfies the Laplace equation. Then

1
Z
0 = △y dy
Ω |x − y|n−2

∂ 1
Z
= dSy .
∂Ω ∂ν(y) |x − y|n−2
(ii) Let x ∈ Ω be fixed, then there is a ball Bρ (x) ⊂ Ω. Then

1
Z
0 = △y dy
Ω\Bρ (x) |x − y|n−2

∂ 1 ∂ 1
Z Z
= dSy − dSy ,
∂Ω ∂ν(y) |x − y|n−2 ∂Bρ (x) ∂ν(y) |x − y|n−2
where in the second integral ν(y) denotes the exterior unit normal at the
boundary of Bρ (x). Using polar coordinates with center at x, we find for
the second integral

∂ 1 ∂
Z Z
ρ2−n ρn−1 dS

n−2
dSy =
∂Bρ (x) ∂ν(y) |x − y| ∂B1 (x) ∂ρ
= (2 − n)ωn .
(iii) Let x ∈ ∂Ω and set for a sufficiently small ρ > 0,
Sρ = Ω ∩ ∂Bρ (x), Cρ = ∂Ω \ Bρ (x),
see Figure 1.4. Then

1
Z
0 = △y dy
Ω\Bρ (x) |x − y|n−2

∂ 1 ∂ 1
Z Z
= dSy − dSy .
Cρ ∂ν(y) |x − y|n−2 Sρ ∂ν(y) |x − y|n−2
Since, see an exercise,

∂ 1 ∂ 1
Z Z
lim dSy = dSy ,
ρ→0 Cρ ∂ν(y) |x − y|n−2 ∂Ω ∂ν(y) |x − y|n−2
Bρ
ν (y) x
Sρ
y ν (y)
y
Ω Cρ
Figure 1.4: Figure to the proof of Lemma 1.2.2
it follows that

∂ 1 ∂ 1
Z Z
dSy = lim dSy .
∂Ω ∂ν(y) |x − y|n−2 ρ→0 Sρ ∂ν(y) |x − y|n−2
We have

∂ 1
Z Z
1−n
dSy = −(n − 2)ρ dSy
Sρ ∂ν(y) |x − y|n−2 Sρ
and
ωn n−1
Z
dSy = ρ 1 + O(ρ2λ ) .
Sρ 2
The previous formula follows by introducing local coordinates at x. Let
(ξ, h(ξ)) ∈ ∂Bρ (x) ∩ ∂Ω, then
|h(ξ)| ≤ cρ1+λ .
Let F be a layer of a sphere with radius ρ of hight cρ1+λ , see Figure 1.5,
then
ωn n−1 ωn n−1
ρ − |F | ≤ |Sρ | ≤ ρ + |F |.
2 2
We have
1 n−1
|F | = ρ ωn (1 − cos θ)
2
1 n−1
= ρ ωn 1 − (1 − c2 ρ2λ )1/2
2
1 n−1
= ρ ωn O(ρ2λ )
2
ρ
δΩ
θ
Figure 1.5: Estimate of |Sρ |
as ρ → 0. ✷
Lemma 1.2.3. Let ∂Ω ∈ C 1,λ . Then

∂ 1
Z

∂ν(y) |x − y|n−2 dSy

∂Ω
is uniformly bounded for x ∈ Rn .
Proof. (i) For fixed d > 0 consider x such that dist(x, ∂Ω) ≥ d/2. Then, see
formula (1.7),
n−1

∂ 1 ≤ (n − 2) 2

∂ν(y) |x − y|n−2 ,
dn−1
which implies that
(n − 2)2n−1

∂ 1
Z

∂ν(y) |x − y|n−2 dSy ≤ |∂Ω|.

∂Ω dn−1
(ii) Consider x ∈ Rn such that dist(x, ∂Ω) < d/2 for a d > 0 and let
x0 ∈ ∂Ω : |x − x0 | = min |x − y|.
y∈∂Ω
Set Sd = ∂Ω ∩ Bd (x0 ). Then for y ∈ ∂Ω \ Sd we have

|y − x| ≥ |y − x0 | − |x − x0 | > d/2,
which implies that
(n − 2)2n−1

∂ 1
Z

dSy ≤ |∂Ω \ Sd |
∂Ω\Sd ∂ν(y)
|x − y|n−2 dn−1
(n − 2)2n−1
≤ |∂Ω|.
dn−1
(iii) Consider
∂ 1
Z

Id := ∂ν(y) |x − y|n−2 dSy .

Sd
In local coordinates, see Figure 1.6, we have, since x − x0 is perpendicular
τ
ν (y)
x0 ξ
y
δΩ
Figure 1.6: Local coordinates, case (iii)
on the tangent plane Tx0 , that x = (0, . . . , 0, δ) and y = (ξ, f (ξ)), where
f ∈ C 1,λ (Dρ ), f (0) = 0, ∇f (0) = 0 and Dρ is a ball in Rn−1 with the center
at 0 ∈ Rn−1 and the radius ρ. We choose d > 0 such that ρ > d. Then, see
formula (1.7),

p
2
∂ 1 ξ · ∇f (ξ) + (δ − f (ξ))
1 + |∇f (ξ)| =
∂ν(y) |x − y|n−2 (|ξ|2 + (δ − f (ξ))2 )n/2
h1 (ξ) + δ
= ,
(|ξ| + (δ + h2 (ξ))2 )n/2
2
where hi ∈ C(Dρ (0)), |hi (ξ)| ≤ c|ξ|1+λ .

Since
|ξ|2 + |δ + h2 (ξ))2 = |ξ|2 + δ 2 + 2δh2 + h22

1
≥ |ξ|2 + δ 2 − 7h22
2
1 2 1 2
≥ |ξ| + δ ,
2 2
provided ρ satisfies 7cρ2λ < 1/2. It follows that
c|ξ|1+λ + |δ|
Z
Id ≤ 2n/2 n/2
dξ. (1.10)
Dρ (|ξ|2 + δ 2 )
The right hand side of (1.10) is uniformly bounded with respect to |δ| < d.
More precisely, we have
Id ≤ 2n/2 ωn−1 max{cλ−1 ρλ , π/2},
see an exercise. ✷
Lemma 1.2.4. Assume σ ∈ C(∂Ω) and x0 ∈ ∂Ω. Then W1 (x), see defini-
tion (1.8), is continuous at x0 .
Proof. Set Sρ = ∂Ω ∩ Bρ (x0 ) and W1 (x) = I1 + I2 , where

∂ 1
Z
I1 (x) = (σ(y) − σ(x0 )) dSy
Sρ ∂ν(y) |x − y|n−2

∂ 1
Z
I2 (x) = (σ(y) − σ(x0 )) dSy .
∂Ω\Sρ ∂ν(y) |x − y|n−2
We have
|W1 (x) − W1 (x0 )| ≤ |I1 (x)| + |I1 (x0 )| + |I2 (x) − I2 (x0 )|
and
∂ 1
Z

|I1 (x)| ≤ |σ(y) − σ(x0 )|

n−2
dSy .
Sρ ∂ν(y) |x − y|
Set, see Lemma 1.2.3,

∂ 1
Z

C = sup ∂ν(y) |x − y|n−2 dSy

x∈Rn ∂Ω
and choose for given ǫ > 0 a ρ = ρ(ǫ) such that

ǫ
|σ(y) − σ(x0 )| <
3C
if y ∈ Sρ . Then |I1 (x)| < ǫ/3 and |I1 (x0 )| < ǫ/3.
Consider x ∈ Rn such that |x − x0 | < ρ/2, then
|y − x| ≥ |y − x0 | − |x − x0 | ≥ ρ/2,
provided that y ∈ ∂Ω \ Sρ . Since I2 is continuous in Bρ/2 (x0 ), there is a

δ = δ(ǫ) such that
|I2 (x) − I2 (x0 )| < ǫ/3
if |x − x0 | < δ(ǫ). Summarizing, we have
|W1 (x) − W1 (x0 )| < ǫ
if |x − x0 | < min{ρ(ǫ)/2, δ(ǫ)}. ✷
Let x0 ∈ ∂Ω and denote by Wi (x0 ) the limit of W (x) from interior to x0

and by We (x0 ) the limit of W (x) from exterior to x0 .
Proposition 1.2.1. Suppose that σ ∈ C(∂Ω) and x0 ∈ ∂Ω. The limits

Wi (x0 ) and We (x0 ) exist and satisfy the jump relations
(n − 2)ωn
Wi (x0 ) = − σ(x0 ) + W (x0 ),
2
(n − 2)ωn
We (x0 ) = σ(x0 ) + W (x0 ).
2
Proof. We will prove the first of the jump relations. For x ∈ Ω we set W (x) =
W1 (x) + σ(x0 )W0 (x), where W1 (x) is continuous at x0 , see Lemma 1.2.4,
and W0 (x) is the Gauss integral, see Lemma 1.2.2. Thus
Wi (x0 ) = lim (W1 (x) + σ(x0 )W0 (x))

x→x0 ,x∈Ω
= W1 (x0 ) − (n − 2)σ(x0 )

∂ 1
Z
= (σ(y) − σ(x0 )) dSy − (n − 2)σ(x0 )
∂Ω ∂ν(y) |x0 − y|n−2
(n − 2)ωn
= W (x0 ) − σ(x0 ).
2
✷
Corollary. The double layer potential

∂ 1
Z
W (x) = σ(y) dSy ,
∂Ω ∂ν(y) |x − y|n−2
where σ ∈ C(∂Ω), defines a solution of the interior Dirichlet problem (Di )

if and only if σ ∈ C(∂Ω) is a solution of the integral equation

(n − 2)ωn ∂ 1
Z
Φ(x) = − σ(x) + σ(y) dSy ,
2 ∂Ω ∂ν(y) |x − y|n−2
where x ∈ ∂Ω, and W (x) is a solution of the exterior Dirichlet problem (De )
if and only if σ ∈ C(∂Ω) satisfies the integral equation

(n − 2)ωn ∂ 1
Z
Φ(x) = σ(x) + σ(y) dSy
2 ∂Ω ∂ν(y) |x − y|n−2
We recall that W is a solution of (Di ) if and only if
Φ(x) = lim W (z),

z→x,z∈Ω
and of (De ) if and only if
Φ(x) = lim W (z).

z→x,z∈Rn \Ω
1.3 Single layer potential

Consider the single layer potential
σ(y)
Z
V (x) = dSy ,
∂Ω |x − y|n−2
where σ ∈ C(∂Ω).
Lemma 1.3.1. V ∈ C(Rn ).
Proof. It remains to show that V (x) is continuous if x ∈ ∂Ω. Let x ∈ ∂Ω,

set Sρ = ∂Ω ∩ Bρ (x), ρ > 0 sufficiently small, and
V (x) = V1 (x) + V2 (x),
where
σ(y)
Z
V1 (x) = dSy ,
Sρ |x − y|n−2
σ(y)
Z
V2 (x) = dSy .
∂Ω\Sρ |x − y|n−2
Consider z ∈ Rn , z in a neighbourhood of x. We have
|V (z) − V (x)| ≤ |V1 (z)| + |V1 (x)| + |V2 (z) − V2 (x)|.

1.3. SINGLE LAYER POTENTIAL 21
ζ ξ
x
y
z δΩ
Figure 1.7: Proof of Lemma 1.3.1
In local coordinates it is y = (ξ, f (ξ)), z = (ζ, δ), where ξ, ζ ∈ Dρ = Dρ (0),

see Figure 1.7, and
p
|σ(ξ, f (ξ))| 1 + |∇f (ξ)|2
Z
|V1 (z)| ≤ (n−2)/2
dξ
Dρ (|ζ − ξ|2 + (δ − f (ξ))2 )
dξ
Z
≤ c
D |ξ − ζ|n−2
Z ρ
dξ
≤ c n−2
D2ρ |ξ|
= 2cωn−1 ρ.
Let ǫ > 0 be given and set ρ = ρ(ǫ) = ǫ/(6cωn−1 ), then |V1 (z)| < ǫ/3 if
|z − x| < ρ(ǫ). Consequently, for those z we have
2
|V (z) − V (x)| ≤ ǫ + |V2 (z) − V2 (x)|.
3
For fixed ρ > 0 there is a δ = δ(ǫ) > 0 such that
ǫ
|V2 (z) − V2 (x)| <
3
if |z − x| < δ(ǫ). Summarizing, we have
|V (z) − V (x)| < ǫ,

provided that |z − x| < min{ρ(ǫ), δ(ǫ)}. ✷
Definition. Assume that u ∈ C 1 (Ω) and ∂Ω ∈ C 1,λ . We say that there

exists a regular interior normal derivative of u at ∂Ω if the limit

∂u(x) ∂u(z)
:= lim
∂ν(x) i z→x ∂ν(x)
exists for each x ∈ ∂Ω. Here is z ∈ Ω on the line defined by the exterior
normal νx at x, see Figure 1.8, and this limit is uniform with respect x ∈ ∂Ω
and it is a continuous function on ∂Ω. Analogously, we define the regular
Bρ
ν (x)
z
ξ
Figure 1.8: Normal derivative
exterior normal derivative of u ∈ C 1 (Rn \ Ω) on ∂Ω by

∂u(x) ∂u(z)
:= lim ,
∂ν(x) e z→x ∂ν(x)
where z ∈ Rn \ Ω is on the line defined by ν(x) and x.
Assume z 6∈ ∂Ω, then

∂V (z) ∂ 1
Z
= σ(y) dSy ,
∂l ∂Ω ∂l |z − y|n−2
where l is any direction. If x ∈ ∂Ω we define

∂V (x) ∂ 1
Z
:= σ(y) dSy . (1.11)
∂ν(x) ∂Ω ∂ν(x) |x − y|n−2
In some of the following considerations we need the formula

n
∂ 1 n−2 X
= − (xi − yi )(ν(x))i . (1.12)
∂ν(x) |x − y|n−2 |x − y|n
i=1
Lemma 1.3.2. The right hand side of (1.11) exists1 if x ∈ ∂Ω.
Proof. We introduce a local coordinate system with center at x as in previous

considerations and show that

∂ 1
Z
Iρ (x) := σ(y) dSy
Sρ ∂ν(x) |x − y|n−2
exists, where Sρ = ∂Ω ∩ Bρ (x), ρ > 0 sufficiently small. In local coordinates

it is y = (ξ, f (ξ)). Using formula (1.12), we obtain
|f (ξ)|
Z
|Iρ (x)| ≤ c1 n
dξ
Dρ |ξ|
Z
≤ c2 |ξ|−n+1+λ dξ
Dρ
= c2 ωn−1 λ−1 ρλ .
Let x ∈ ∂Ω and consider the sum

∂V (z)
s(z) = + W (z)
∂ν(x)

∂ 1 ∂ 1
Z
= σ(y) + dSy ,
∂Ω ∂ν(x) |z − y|n−2 ∂ν(y) |z − y|n−2
where W is the dipole potential and z is on the line defined by ν(x), see
Figure 1.8.
Lemma 1.3.3. The sum s(z) is continuous at x.
Proof. Set Sρ = ∂Ω ∩ Bρ (x), ρ > 0 sufficiently small, and
s(z) = s1 (z) + s2 (z),

1
i. e., this weakly singular integral exists in the sense of Riemann or as a Lebesgue
integral, and it is bounded
where

∂ 1 ∂ 1
Z
s1 (z) = σ(y) + dSy ,
Sρ ∂ν(x) |z − y|n−2 ∂ν(y) |z − y|n−2

∂ 1 ∂ 1
Z
s2 (z) = σ(y) + dSy .
∂Ω\Sρ ∂ν(x) |z − y|n−2 ∂ν(y) |z − y|n−2
We have
|s(z) − s(x)| ≤ |s1 (z)| + |s1 (x)| + |s2 (z) − s2 (x)|.
The lemma is shown if for given ǫ > 0 there exists a ρ = ρ(ǫ) > 0 such
that |s1 (z)| < ǫ/3 if |x − z| < ρ(ǫ), see the proof of Lemma 1.3.1. We have,
see (1.7), (1.12),

∂ 1 ∂ 1
+
∂ν(x) |z − y|n−2 ∂ν(y) |z − y|n−2
n n
!
1 X X
= (n − 2) (zi − yi )(ν(y))i − (zi − yi )(ν(x))i ,
|z − y|n
i=1 i=1
where, in local coordinates, x = (0, . . . , 0, 0), z = (0, . . . , 0, δ), ν(x) =

(0, . . . , 0, 1) and
1
ν(y) = p (−fξ1 , . . . , −fξn−1 , 1).
1 + |∇f (ξ)|2
It follows that

∂ 1 ∂ 1
Z

|s1 (z)| ≤ c1 ∂ν(x) |z − y|n−2 + ∂ν(y) |z − y|n−2 dSy

Sρ
p
ξ · ∇f (ξ) + (δ − f (ξ)) − δ 1 + |∇f (ξ)|2
Z
≤ c2 dξ
Dρ (0) (|ξ|2 + |δ − f (ξ)|2 )n/2
|ξ|1+λ + |δ||ξ|2λ
Z
≤ c3 dξ
Dρ (0) (|ξ|2 + δ 2 )n/2
≤ c3 ωn−1 max{λ−1 ρλ , πρ2λ /2},
where the constants ci are independent of ρ. ✷
Proposition 1.3.1. Suppose that ∂Ω ∈ C 1,λ . Then there exists a regular

interior and a regular exterior normal derivative of V , and these derivatives
satisfy the jump relations

∂V (x) (n − 2)ωn ∂V (x)
= σ(x) +
∂ν(x) i 2 ∂ν(x)

∂V (x) (n − 2)ωn ∂V (x)
= − σ(x) + ,
∂ν(x) e 2 ∂ν(x)
where x ∈ ∂Ω.
Proof. The existence of regular normal derivatives follow from Lemma 1.3.3
and Proposition 1.2.1 since

∂V (z) ∂V (z)
= + W (z) − W (z),
∂ν(x) ∂νx
where z is on the line defined by ν(x). From Lemma 1.3.3 it follows

∂V (x) ∂V (x)
+ Wi (x) = + We (x)
∂ν(x) i ∂ν(x) e
∂V (x)
= + W (x),
∂ν(x)
where

∂V (x) ∂ 1
Z
: = σ(y) dSy ,
∂ν(x) ∂ν(x) |x − y|n−2
Z∂Ω
∂ 1
W (x) : = σ(y) dSy .
∂Ω ∂ν(y) |x − y|n−2
Using Lemma 1.3.1, we obtain

∂V (x) ∂V (x)
= W (x) − Wi (x) +
∂ν(x) i ∂ν(x)
(n − 2)ωn ∂V (x)
= σ(x) +
2 ∂ν(x)
and

∂V (x) ∂V (x)
= W (x) − We (x) +
∂ν(x) e ∂νx
(n − 2)ωn ∂V (x)
= − σ(x) + .
2 ∂ν(x)
✷
Remark. Let x ∈ ∂Ω, then it follows immediately that

∂V (x) ∂V (x)
− = (n − 2)ωn σ(x).
∂ν(x) i ∂νx e
Corollary. The single layer potential

σ(y)
Z
V (x) = dSy ,
∂Ω |x − y|n−2
where σ ∈ C(∂Ω) defines a solution of the interior Neumann problem (Ni )
if and only if σ ∈ C(∂Ω) is a solution of the integral equation

(n − 2)ωn ∂ 1
Z
Ψ(x) = σ(x) + σ(y) dSy ,
2 ∂Ω ∂ν(x) |x − y|n−2
where x ∈ ∂Ω, and V (x) is a solution of the exterior Neumann problem (Ne )
if and only if σ ∈ C(∂Ω) satisfies the integral equation

(n − 2)ωn ∂ 1
Z
Ψ(x) = − σ(x) + σ(y) dSy .
2 ∂Ω ∂ν(x) |x − y|n−2
We recall that V is a solution of (Ni ) if and only if

∂V (x)
Ψ(x) = ,
∂ν(x) i
and of (Ne ) if and only if

∂V (x)
Ψ(x) = .
∂ν(x) e
1.4 Integral equations

Denote by H = L2 (∂Ω) the Hilbert space with the inner product
Z
hσ, µi = σ(x)µ(x) dSx ,
∂Ω
and, if n ≥ 3, we define the linear operator T from H into H by

∂ 1
Z
(T σ)(x) = σ(y) dSy .
∂Ω ∂ν(y) |x − y|n−2
1.4. INTEGRAL EQUATIONS 27
Set
∂ 1
Z
∗
(T σ)(x) = σ(y) dSy ,
∂Ω ∂ν(x) |x − y|n−2
then
hT σ, µi = hσ, T ∗ µi
for all σ, µ ∈ H. Below we will show that T is bounded. Then it follows

that T ∗ is the adjoint operator to T .
According to the above corollaries to Proposition 1.2.1 and Proposi-
tion 1.3.1, the potentials W and V are solutions of the boundary value
problems (Di ), De ), (Ni ) and (Ne ) if the density σ is continuous on ∂Ω and
satisfies the integral equations, respectively,
2 2
σ− Tσ = − Φ (Di )I
(n − 2)ωn (n − 2)ωn
2 2
σ+ Tσ = Φ (De )I
(n − 2)ωn (n − 2)ωn
2 2
σ+ T ∗σ = Ψ (Ni )I
(n − 2)ωn (n − 2)ωn
2 2
σ− T ∗σ = − Ψ (Ne )I .
(n − 2)ωn (n − 2)ωn
Remark. Since we make the ansatz with above potentials for the exterior
problems, we prescribe in fact the behavior |u(z) ≤ c|z|1−n , |u(z) ≤ c|z|2−n ,
respectively, as z → ∞.
The above integral equations are defined for σ ∈ L2 (∂Ω). In the following we
will discuss whether or not there exist solutions in L2 (∂Ω). From a regularity
result which says that an L2 -solution is in fact in C(∂Ω), we recover that the
potentials define solutions of the boundary value problem, see the corollaries
to Proposition 1.2.1 and Proposition 1.3.1.
Proposition 1.4.1. Suppose that ∂Ω ∈ C 1,λ . Then T is a completely

continuous operator from H into H.
Proof. (i) T is bounded. It is sufficient, see Section 1.1, to show that

Z
(P µ)(ζ) := a(ξ)µ(ξ)K(ξ, ζ) dξ
Dρ
is bounded from L2 (Dρ ) into L2 (Dρ ). Here is Dρ = Dρ (0) ⊂ Rn−1 , a ∈

C0∞ (Dρ ), µ(ξ) = σ(ξ, f (ξ)) and
(n − 2) (−(ξ − ζ) · ∇f (ξ) + f (ζ) − f (ξ))

K(ξ, ζ) = .
(|ξ − ζ|2 + (f (ξ) − f (ζ))2 )n/2
Set q(ζ) := (P µ)(ζ), then
A(ξ, ζ)
Z
q(ζ) = µ(ξ) dξ,
Dρ |ξ − ζ|n−1−λ
where A is bounded on Dρ × Dρ ) and continuous if ξ 6= ζ. Let κ = n − 1 − λ,

then we have, with constants ci independent of µ and ρ, that
|µ(ξ)| 1
Z
|q(ζ)| ≤ c1 κ/2 |ξ − ζ|κ/2
dξ,
Dρ |ξ − ζ|
|µ(ξ)|2 dξ
Z Z
2
|q(ζ)| ≤ c2 κ
dξ
Dρ |ξ − ζ| Dρ |ξ − ζ|κ
|µ(ξ)|2
Z
≤ c 3 ρλ κ
dξ,
Dρ |ξ − ζ|
!
dζ
Z Z Z
2 λ 2
|q(ζ)| dζ ≤ c3 ρ |µ(ξ)| κ
dξ
Dρ Dρ Dρ |ξ − ζ|
Z
2λ
≤ c4 ρ |µ(ξ)|2 dξ.
Dρ
(ii) T is completely continuous. According to a lemma due to Kolmogoroff,

see for example [21], pp. 246, or [1], pp. 31, P is completely continuous if
for given ǫ1 > 0 there exists an h0 (ǫ1 ) > 0 such that
Z
|q(ζ + h) − q(ζ)|2 dζ ≤ ǫ21
Dρ
for all h ∈ Rn−1 such that |h| ≤ h0 (ǫ1 ), and uniformly for ||µ||L2 (Dρ ) ≤ M ,
where M < ∞. Thus, the set ||µ||L2 (Dρ ) ≤ M is uniformly continuous in the
mean. Above we set q(ζ) = 0 if ζ 6∈ Dρ . Let η ∈ C(R+ ), 0 ≤ η ≤ 1, such
that for given ǫ > 0

1 : 0 ≤ t ≤ ǫ/2
η(t) = .
0 : t≥ǫ
Set
q(ζ) = q1 (ζ) + q2 (ζ),
where
A(ξ, ζ)η(|ξ − ζ|)

Z
q1 (ζ) = µ(ξ) dξ
Dρ (0) |ξ − ζ|κ
A(ξ, ζ) (1 − η(|ξ − ζ|))
Z
q2 (ζ) = µ(ξ) dξ.
Dρ (0) |ξ − ζ|κ
We have
|q(ζ + h) − q(ζ)| ≤ |q1 (ζ + h)| + |q1 (ζ)| + |q2 (ζ + h) − q2 (ζ)|. (1.13)
Let ǫ > 0 be fixed, then for given τ > 0 there is an h0 = h0 (τ ) > 0 such that
|q2 (ζ) − q2 (ζ)| < τ (1.14)
for all |h| ≤ h0 . Concerning q1 (ζ) we have
dξ
Z
|q1 (ζ)| ≤ c |µ(ξ)|
D (0)∩Bǫ (ζ) |ξ − ζ|κ
Zρ
dξ
|q1 (ζ)|2 ≤ c2 |µ(ξ)|2
Dρ (0)∩Bǫ (ζ) |ξ − ζ|κ
dξ
Z
κ
Dρ (0)∩Bǫ (ζ) |ξ − ζ|
dξ
Z
≤ c2 ωn−1 ǫλ λ−1 |µ(ξ)|2
D (0)∩B (ζ) |ξ − ζ|κ
Z ρ Zǫ
dξ
Z
|q1 (ζ)|2 dζ ≤ c2 ωn−1 ǫλ λ−1 |µ(ξ)|2 dζ
Dρ (0) Dρ (0) Dρ (0) |ξ − ζ|κ
dζ
Z Z
2 λ −1 2
= c ωn−1 ǫ λ |µ(ξ)| dξ κ
D (0) D (0) |ξ − ζ|
Z ρ Z ρ
dζ
≤ c2 ωn−1 ǫλ λ−1 |µ(ξ)|2 dξ κ
Dρ (0) D2ρ (ξ) |ξ − ζ|
= c2 ωn−1
2
ǫλ λ−2 (2ρ)λ ||µ||2L2 (Dρ ) .
Analogously, we have
dξ
Z
|q1 (ζ + h)|2 ≤ c2 |µ(ξ)|2
Dρ (0)∩Bǫ (ζ+h) |ξ − (ζ + h)|κ
dξ
Z
Dρ (0)∩Bǫ (ζ+h) |ξ − (ζ + h)|κ

dξ
Z
2 λ −1
≤ c ωn−1 ǫ λ |µ(ξ)|2
D (0)∩B (ζ+h) |ξ − (ζ + h)|κ
Z ρ Zǫ
dξ
Z
|q1 (ζ + h)|2 dζ ≤ c2 ωn−1 ǫλ λ−1 |µ(ξ)|2 dζ
Dρ (0) Dρ (0) Dρ (0) |ξ − (ζ + h)|κ
dζ
Z Z
= c2 ωn−1 ǫλ λ−1 |µ(ξ)|2 dξ
D (0) Dρ (0) |ξ − (ζ + h)|κ
Z ρ
≤ c2 ωn−1 ǫλ λ−1 |µ(ξ)|2 dξ
Dρ (0)
dζ
Z
κ
D3ρ (ξ−h) |ζ − (ξ − h)|
= c2 ωn−1
2
ǫλ λ−2 (3ρ)λ ||µ||2L2 (Dρ ) .
Combining these L2 -estimates with (1.13) and (1.14) we obtain that the
mapping T is completely continuous. ✷
From a result of functional analysis we have
Corollary. T ∗ is bounded with the same norm as T and T ∗ is completely

continuous.
In the following we study the question of the existence of solutions σ ∈

L2 (∂Ω) of the above integral equations. To recover that the associated
surface potentials define solutions of the original boundary value problems,
we need more regularity, namely σ ∈ C(∂Ω). We obtain this property by
using the integral equations.
Proposition 1.4.2 (Regularity). Let w ∈ L2 (Dρ ) be a solution of the

integral equation
A(ξ, ζ)
Z
w(ζ) − w(ξ) dξ = b(ζ),
Dρ |ξ − ζ|κ
where κ = n − 1 − λ, Dρ = Dρ (0) ⊂ Rn−1 , A is bounded in Dρ × Dρ and

continuous if ξ 6= ζ. The function b ∈ C(Dρ ) is given. Then it follows that

w ∈ C(Dρ ).
Proof. Let η ∈ C(R+ ), 0 ≤ η ≤ 1, such that for given ǫ > 0

1 : 0 ≤ t ≤ ǫ/2
η(t) = .
0 : t≥ǫ
Set
A(ξ, ζ)
= K1 (ξ, ζ) + K2 (ξ, ζ),
|ξ − ζ|κ
where
A(ξ, ζ)η(|ξ − ζ|)
K1 (ξ, ζ) =
|ξ − ζ|κ
A(ξ, ζ) (1 − η(|ξ − ζ|)
K2 (ξ, ζ) = .
|ξ − ζ|κ
Then Z
w(ζ) − w(ξ)K1 (ξ, ζ) dξ = g(ζ),
Dρ
where Z
g(ζ) = w(ξ)K2 (ξ, ζ) dξ + b(ζ)
Dρ
is a continuous function on Dρ . Define the integral operator T1 from L2 (Dρ )

into L2 (Dρ ) by Z
(T1 w)(ζ) = w(ξ)K1 (ξ, ζ) dξ,
Dρ
then we can write the above integral equation as (I − T1 )w = g, where I

denotes the identity operator. The L2 -norm of T1 satisfies the inequality
||T1 || < 1, provided ǫ > 0 is sufficiently small, see an exercise. It follows
that w is given by the Neumann series
∞
X
w = (I − T1 )−1 g = T1n g,
n=0
which is a uniformly convergent series of continuous functions, provided

ǫ > 0 was chosen sufficiently small, see an exercise. ✷
FREDHOLM THEOREMS. Here we recall some results from functional

analysis, see for example [23]. Let H be a Hilbert space over C and T : H 7→
H a completely continuous linear operator. Consider for given f, g ∈ H

and λ ∈ C the equations
u + λ Tu = f (I)
∗
v+λ T v = g (I ∗ )
and the associated homogeneous equations
u + λ Tu = 0 (Ih )
∗
v+λ T v = 0 (Ih∗ ).
Equations (I ∗ ), (Ih∗ ) are called adjoint to (I), (Ih ), respectively.
(i) Let λ be an eigenvalue of (Ih ), then the linear space of solutions has
finite dimension.
(ii) The eigenvalue problem (Ih ) has at most a countable set of eigenvalues
with at most one limit element at infinity.
(iii) λ is an eigenvalue of (Ih ) if and only if λ is an eigenvalue of (Ih∗ ) and

dim N (I + λT ) =dim N (I + λT ∗ ).
(iv) (I) has a solution if and only if f ⊥ N (I +λ T ∗ ) and (I ∗ ) has a solution

if and only if g ⊥ N (I + λ T ).
We recall that N (A) denotes the null space N (A) = {w ∈ H : Aw = 0} of

a linear operator A.
Proposition 1.4.3. Suppose that ∂Ω ∈ C 2 , then λ0 = −2/((n − 2)ωn ) is

no eigenvalue of the homogeneous integral equation to (Di ).
Proof. Suppose that λ0 is an eigenvalue and µ0 ∈ L2 (∂Ω) an associated

eigenvector of the adjoint problem (Ne )I . From Proposition 1.4.2 we have
that µ0 ∈ C(∂Ω). Consider the single layer potential
µ0 (y)
Z
V (x) := n−2
dSy .
∂Ω |x − y|
From a jump relation of Proposition 1.3.1 it follows that

∂V (x)
=0 (1.15)
∂ν(x) e
since µ0 is an eigenvector of (Ne )I .

Set for a (small) h > 0
Ωh = Ω ∪ {y ∈ Rn : y = x + sν(x), x ∈ ∂Ω, 0 ≤ s < h},
see Figure 1.9. The surface ∂Ωh is called parallel surface to ∂Ω.
ν (x)= ν(z)
h
x
δΩ h
Figure 1.9: Parallel surface
Consider a ball BR = BR (0) such that Ωh ⊂ BR , then
∂V (x) ∂V (z)
Z Z Z
2
|∇V | dx = V (x) dSx − V (z) dSz . (1.16)
BR \Ωh ∂BR ∂ν(x) ∂Ωh ∂ν(z)
We have on ∂Ωh
ν(z) = ν(x). (1.17)
To show this equation, we consider the surface ∂Ω which is given (locally)
by x = x(u), where u ∈ U and U is an (n–1)-dimensional parameter do-
main. The parallel surface ∂Ωh is defined by z(u) = x(u) + hν(x(u)). Then
we consider a C 1 -curve X(t) on ∂Ω with X(0) = x, and let Z(t) be the
associated curve on ∂Ωh . Then
|X(t) − Z(t)|2 = h2 .
It follows
(X(t) − Z(t)) · X ′ (t) − (X(t) − Z(t)) · Z ′ (t) = 0,
which proves (1.17) since the first term is zero.

Combining (1.16), (1.17), (1.15) and

∂V (z) ∂V (x)
lim = ,
h→0 ∂ν(x) ∂ν(x) e
we obtain
∂V (x)
Z Z
2
lim |∇V | dx = V (x) dSx
h→0 BR \Ωh ∂BR ∂ν(x)
since the surface element dSz converges uniformly to dSx on U as h → 0.2

We have V = O(R2−n ) and ∂V /∂ν(x) = O(R1−n ), consequently
Z !
lim lim |∇V |2 dx = 0.
R→∞ h→0 BR \Ωh
Thus V = const. on Rn \ Ω. From the behavior of V at infinity it follows

that V ≡ 0 on Rn \ Ω. Because of V ∈ C(Rn ), see Lemma 1.3.1.
From the maximum principle we find that V ≡ 0 in Ω since △V = 0 in
Ω and V = 0 on ∂Ω. Consequently, the interior regular normal derivative
on ∂Ω is zero. Finally the jump relations, see Proposition 1.3.1, imply that
µ0 (x) ≡ 0 on ∂Ω. ✷
Proposition 1.4.3 and Fredholm’s theorems imply
Theorem 1.4.1. Let Ω be bounded and ∂Ω ∈ C 2 . Then there exists for

given Φ, Ψ ∈ C(∂Ω) a unique solution of the interior Dirichlet problem (Di )
and the exterior Neumann problem (Ne ), respectively.3
Proof. N (I + λ0 T ) = N (I + λ0 T ∗ ) = {0}. ✷
Proposition 1.4.4. Let Ω be bounded and ∂Ω ∈ C 2 . The number λ0 =

2/((n − 2)ωn ) is a simple eigenvalue of (De )I to the eigenvector σ ≡ 1.
Proof. From, see Lemma 1.2.2,

∂ 1 (n − 2)ωn
Z
n−2
dSy = −
∂Ω ∂ν(y) |x − y| 2
2
√
In the case R3 we have dSz = EG − F 2 du1 du2 , where E = zu1 · zu1 , G = zu2 · zu2 ,
F = zu1 · zu2 and z(u) = x(u) + hν(x(u)).
3
In this and in the following two theorems it is sufficient to assume that ∂Ω ∈ C 1,1 by
Rademacher’s Theorem: A Lipschitz continuous function is differentiable almost every-
where, see for example [5], pp. 280 or [6].
we see that λ0 is an eigenvalue and σ ≡ 1 is an associated eigenvector. From

Fredholm’s theorems it follows that there exists at least one eigenfunction
µ0 (x) to λ0 of (Ni )h . We will show that dim N (I + λ0 T ∗ ) = 1. Set
µ0 (y)
Z
V (x) = dSy .
∂Ω |x − y|n−2
From the jump relations, see Proposition 1.3.1 and from the fact that µ0
is an eigenvector it follows that (∂V (x)/∂ν(x))i = 0. We obtain as in the
proof of Proposition 1.4.3 that V (x) = const. =: c0 in Ω. This constant is
different from zero. If not, then V = 0 on ∂Ω. Then the maximum principle
implies that V ≡ 0 in Rn \ Ω. We recall that V = O(|x|2−n ) as |x| → ∞.
Consequently, we have (∂V (x)/∂ν(x))e = 0, which implies that µ0 = 0, see
the jump relations of Proposition 1.3.1.
Let µ1 be another eigenvector to λ0 , then we can assume that µ1 ∈ C(∂Ω)
according to the regularity result Proposition 1.4.2. Set
µ1 (y)
Z
V1 (x) = n−2
dSy .
∂Ω |x − y|
As above, we conclude that V1 (x) ≡ const. =: c1 in Ω, where c1 6= 0.

The linear combination µ2 := c1 µ0 − c0 µ1 is contained in the null space
N (I + λ0 T ∗ ). Set
µ2 (y)
Z
V2 (x) = n−2
dSy
∂Ω |x − y|
= c1 V0 (x) − c0 V1 (x).
In Ω we have V2 (x) = c1 c0 − c0 c1 , and from the jump relations we find as

above that µ2 (x) ≡ 0, then
c1
µ1 (x) = µ0 (x).
c0
✷
From Fredholm theorems it follows
Theorem 1.4.2. Let Ω be bounded and ∂Ω ∈ C 2 . Then there exists a

solution of (Ni ) if and only if
Z
Ψ(y) dSy = 0.
∂Ω
In fact, we obtain also the existence of a solution of (De )I under the as-
sumption that Z
µ0 (y)Φ(y) dSy = 0,
∂Ω
where µ0 is the eigenvector from above. It turns out that there is a solution of
the exterior Dirichlet problem without this restriction if we look for solutions
with a weaker decay at infinity. We make the ansatz of a sum of a double
layer and a single layer potential

∂ 1
Z
u(x) = σ(y) dSy
∂Ω ∂ν(y) |x − y|n−2
µ0 (y)
Z
+d dSy ,
∂Ω |x − y|n−2
where d is a constant which we will determine later. The ansatz defines a
solution of the exterior Dirichlet problem if and only if
lim u(y) = φ(x),

y→x, y∈Rn \Ω
where x ∈ ∂Ω. From a jump relation of Proposition 1.2.1 we see that the
unknown density σ must satisfy the integral equation

∂ 1
Z
Φ(x) = σ(y) dSy
∂Ω ∂ν(y) |x − y|n−2
(n − 2)ωn µ0 (y)
Z
+ σ(x) + d n−2
dSy .
2 ∂Ω |x − y|
Above we have shown that, if x ∈ ∂Ω,

µ0 (y)
Z
n−2
dSy = const. = c0
∂Ω |x − y|
with a constant c0 6= 0. Thus we have to consider the integral equation

(n − 2)ωn ∂ 1
Z
σ(x) + σ(y) dSy = Φ(x) − dc0 .
2 ∂Ω ∂ν(y) |x − y|n−2
This equation has a solution if
Z
(Φ(x) − dc0 )µ0 (x) dSx = 0
∂Ω
is satisfied. We can find an appropriate constant d such that this equation

is satisfied since Z
µ0 (x) dSx 6= 0.
∂Ω
This inequality is a consequence of a jump relation and of the fact that µ0
is an eigenvector:

∂V0 (x) (n − 2)ωn ∂ 1
Z
= − µ0 (x) + µ0 (y) dSy
∂ν(x) e 2 ∂Ω ∂ν(x) |x − y|n−2

(n − 2)ωn ∂ 1
Z
0 = µ0 (x) + µ0 (y) dSy ,
2 ∂Ω ∂ν(x) |x − y|n−2
which implies that

∂V0 (x)
= −(n − 2)ωn µ0 (x).
∂ν(x) e
Suppose that Z
µ0 (x) dSx = 0,
∂Ω
then
∂V0 (x)
Z
dSx = 0,
∂Ω ∂ν(x) e
which implies, see the proof of Proposition 1.4.3 for notations,
∂V0 (x) ∂V0 (z)
Z Z Z
|∇V0 |2 dx = V0 (x) dSx − V0 (z) dSz .
BR \Ωh ∂BR ∂ν(x) ∂Ωh ∂ν(z)
Letting h → 0 and R → ∞, it follows V0 = const. in Rn \ Ω and V0 = c0

since V0 (x) = const. = c0 on ∂Ω.
From the decay behaviour of V0 at infinity and since V ∈ C(Rn ) we
find that V0 = 0 in Rn \ Ω. Thus we have c0 = 0, a contradiction to a
consideration above.
Thus, we have shown
Theorem 1.4.3. Let Ω be bounded and ∂Ω ∈ C 2 . Then for given Φ ∈

C(∂Ω) there exists a unique solution u of (De ) with the property u = O(|x|2−n )
as |x| → ∞.
Proof. It remains to show that the solution is unique. This follows from the
maximum principle. ✷
Remark. There is no uniqueness without the decay assumption. Let Ω =

BR (0) be a ball in R3 . Then u = 1/|x| and u = 1/R are two solutions of the
Laplace equation with the same boundary values on ∂Ω.
1.5 Volume potential

Set
Γ(x, y) = s(|x − y|),
where s(r) is the singularity function
1
(
− 2π ln r : n=2
s(r) := r 2−n
(n−2)ωn
: n≥3
We recall that ωn = |∂B1 (0)|. Let Ω ∈ Rn be a bounded and sufficiently

regular domain. We define for given f the volume potential (or Newton
potential) Z
V (x) = Γ(x, y)f (y) dy.
Ω
If f is bounded in Ω and f ∈ C 1 (Ω), then V ∈ C 2 (Ω) and −△V = f in

Ω, see for example [17]. This result holds under the weaker assumption
that f is bounded and locally Hölder continuous in Ω, see Proposition 1.5.1.
In fact, also the second derivatives are Hölder continuous (with the same
Hölder exponent), see [10], for example.
Definition. Let f be a real function, defined in a fixed bounded neighbor-

hood D of x0 ∈ Rn . Then f is called Hölder continuous at x0 if there exists
a real number α, 0 < α ≤ 1, such that
|f (x) − f (x0 )|
[f ]α,x0 := sup < ∞.
x∈D\{x0 } |x − x0 |α
The constant [f ]α,x0 is called Hölder constant and α Hölder exponent.

The function f is said to be uniformly Hölder continuous with respect
to x, y in D if
|f (x) − f (y)|
[f ]α,D := sup < ∞,
x,y∈D, x6=y |x − y|α
and f is called locally Hölder continuous in a domain Ω if f is uniformly

Hölder continuous on compact subsets of Ω.
1.5. VOLUME POTENTIAL 39
In the following we will use the abbreviations Di = ∂/∂xi and Dij =

∂ 2 /∂xi ∂xj .
Proposition 1.5.1. (i) Let f be bounded and integrable over Ω. Then

V ∈ C 1 (Rn ) and for any x ∈ Ω
Z
Di V (x) = Di Γ(x, y)f (y) dy.
Ω
(ii) Let f be bounded and locally Hölder continuous in Ω with exponent 0 <
α ≤ 1. Then V ∈ C 2 (Ω), −△V = f in Ω, and for any x ∈ Ω
Z
Dij V (x) = Dij Γ(x, y)(f (y) − f (x)) dy (1.18)
Ω0
Z
−f (x) Di Γ(x, y)(ν(y))j dSy ,
∂Ω0
where Ω0 ⊃ Ω is any domain for which the divergence theorem holds and f
is extended to vanish outside Ω.
Proof. See [10], Chapter 4. (i) Set for x ∈ Rn

Z
v(x) = Di Γ(x, y)f (y) dy.
Ω
This function is well defined since |Di Γ| ≤ |x − y|1−n /ωn holds. We will
show that v = Di V and v ∈ C(Rn ). Let η ∈ C 1 (R) be a fixed function
satisfying 0 ≤ η ≤ 1, 0 ≤ η ′ ≤ 2, η(t) = 0 if t ≤ 1 and η(t) = 1 if t ≥ 2. For
a (small) ǫ > 0 set ηǫ = η(|x − y|/ǫ) and consider the regularized potential
Z
Vǫ (x) = Γ(x, y)ηǫ f (y) dy.
Ω
Then Vǫ ∈ C 1 (Rn ) and

Z
v(x) − Di Vǫ (x) = Di ((1 − ηǫ )Γ) f (y) dy.
B2ǫ (x)
We obtain
Z
|v(x) − Di Vǫ (x)| ≤ sup |f | (|Di Γ| + 2|Γ|/ǫ) dy
Ω B2ǫ (x)

4(1 + | ln(2ǫ)|)ǫ : n=2
≤ sup |f | .
Ω 2nǫ/(n − 2) : n≥3
It follows that Vǫ and Di Vǫ converge uniformly on compact subsets of Rn to

V and v, resp., as ǫ → 0. Consequently V ∈ C 1 (Rn ) and Di V = v.
(ii) Set for x ∈ Ω

Z
u(x) = Dij Γ(x, y)(f (y) − f (x)) dy
Ω0
Z
−f (x) Di Γ(x, y)(ν(y))j dSy .
∂Ω0
The right hand side is well-defined since f is locally Hölder continuous and
since |Dij Γ| ≤ (1 + n)|x − y|n /ωn holds. Set v = Di V and define for ǫ > 0
the regularized function
Z
vǫ (x) = Di Γηǫ f (y) dy.
Ω0
Then
Z
Dj vǫ = Dj (Di Γηǫ )f (y) dy
Ω0
Z
= Dj (Di Γηǫ )(f (y) − f (x)) dy
Ω0
Z
f (x) Dj (Di Γηǫ ) dy
Ω0
Z
= Dj (Di Γηǫ )(f (y) − f (x)) dy
Ω0
Z
−f (x) Di Γ(ν(y))j dSy ,
∂Ω0
provided ǫ > 0 is small enough such that ηǫ = 1 on ∂Ω0 , see Figure 1.10.
Then
Z
u(x) − Dj vǫ (x) = Dj ((1 − ηǫ )Di Γ) (f (y) − f (x)) dy.
B2ǫ (x)
We suppose that 2ǫ < dist (x, ∂Ω) if x ∈ Ωc , Ωc ⊂⊂ Ω. Then

Z
|u(x) − Dj vǫ (x)| ≤ [f ]α,Ωc (|Dij Γ| + 2|Di Γ|/ǫ) |x − y|α dy
B2ǫ (x)
≤ [f ]α,Ωc (4 + n/α)(2ǫ)α .
1.5. VOLUME POTENTIAL 41
δΩ 0
Ωc x
y
δΩ
Figure 1.10: Proof of Proposition 1.5.1
It follows that V ∈ C 2 (Ω) and u = Dij V since Di vǫ converges to u uniformly

on compact subsets of Ω.
Set in formula (1.18) Ω0 = BR (x), R sufficiently large, then
Z n
X
△V = −f (x) Di Γ(x, y)(ν(y))i dSy
∂BR (x) i=1
1
Z
= −f (x) ν(y) · ν(y) dSy
ωn Rn−1 ∂BR (x)
= −f (x).
From Proposition 1.5.1 and Theorem 1.4.1 we obtain
Theorem 1.5.1. Let Ω be a bounded domain with ∂Ω ∈ C 2 , f be bounded

and locally Hölder continuous in Ω and Φ ∈ C(∂Ω). Then there exists a
unique solution u ∈ C 2 (Ω) ∩ C(Ω) of the Dirichlet problem −△u = f in Ω,
u = Φ on ∂Ω.
Proof. Set u = V + w, where

Z
V (x) = Γ(x, y)f (y) dy.
Ω
Then u is a solution of the Dirichlet problem if and only if △w = 0 in Ω and

w = Φ − V on ∂Ω. The existence of a w follows from Theorem 1.4.1. The
uniqueness of u is a consequence of the maximum principle. Moreover, w is
given by a dipole potential, see Theorem 1.4.1. ✷
1.6. EXERCISES 43
1.6 Exercises
1. Let S be a surface in R3 defined by z = f (x), x = (x1 , x2 ), x ∈ U ,
where U is a neighborhood of x = 0. Assume f ∈ C 1,λ (U ), f (0) =
0, ∇f (0) = 0. Consider the intersection I = S ∩ ∂Bρ0 (0), ρ0 > 0
sufficiently small, i. e.,
I = {(x, z) : z = f (x) and x21 + x22 + z 2 = ρ20 }.
Show that there is a function ǫ(ρ, φ), 0 < ρ ≤ ρ0 , φ ∈ [0, 2π), 2π-
periodic in φ and in C 1 with respect to φ, such that ǫ(ρ, φ) = O(ρλ )
as ρ → 0, uniformly in φ ∈ [0, 2π), and
x1 = ρ(1 + ǫ(ρ, φ)) cos φ

x2 = ρ(1 + ǫ(ρ, φ)) sin φ.
Hint: Implicit function theorem.
2. Let B2R (x0 ) ⊂ Rn be a ball with radius 2R and the center at x0 . Show
that there is a function η ∈ C0∞ (B2R (x0 )) which satisfies 0 ≤ η ≤ 1 in
B2R (x0 ) and η ≡ 1 in BR (x0 ).
Hint: Set r = |x| and define

2 −r 2 /4)
φ(r) = e−1/((3/(2R)−r)
if R < r < 2R and set φ(r) = 0 if 0 < r < R or r ≥ 2R. Let

Rr
0 φ(t) dt
ψ(r) = R ∞
0 φ(t) dt
and χ(r) = 1 − ψ(r). Show that η(x) = χ(|x − x0 |) is a function which

satisfies the above properties.
3. Suppose that ∂Ω ∈ C 1,λ and x ∈ ∂Ω. Show that

∂ 1
Z
lim dSy = 0
ρ→0 ∂Ω\Bρ (x) ∂ν(y) |z − y|n−2
Hint: Local coordinates and formula (1.7).

4. Let Ω ⊂ Rn be a bounded and sufficiently regular domain and set
A(ξ, ζ)
Z
q(ζ) = dξ,
Ω |ξ − ζ|n−λ
where A is bounded in Ω × Ω and continuous if ξ 6= ζ and 0 < λ ≤ 1.

Show that q ∈ C(Ω).
Hint: Let η ∈ C(R) be a fixed function satisfying 0 ≤ η ≤ 1, η(t) = 0
if t ≤ 1 and η(t) = 1 if t ≥ 2. For (small) ǫ > 0 set ηǫ = η(|ζ − ξ|/ǫ).
Then consider the regularized function
A(ξ, ζ)
Z
qǫ (ζ) = ηǫ dξ
Ω |ξ − ζ|n−λ
and prove that qǫ converges uniformly to q in Ω.
5. Show that
||K1 w||L2 (Dρ ) ≤ cǫλ ||w||L2 (Dρ ) .
For the definition of K1 see the proof of Proposition 1.4.2.
6. Let g ∈ C(Dρ ). Prove that
(i) |K1l g| ≤ (cǫλ )l maxDρ |g(ζ)|.
(ii) K1l g are continuous on Dρ .
(iii) ∞ l
P
l=1 K1 g is uniformly convergent on Dρ , provided that ǫ > 0 is
small enough.
For the definition of K1 see the proof of Proposition 1.4.2.
7. The solution u of the interior Dirichlet problem △u = 0 in Ω and

u = Φ on ∂Ω, where Ω ⊂ R2 and Φ ∈ C(∂Ω), is given by
∂
Z
u(x) = − σ(y) ln(|x − y|) dSy .
∂Ω ∂ν(y)
Here is σ(x), x ∈ ∂Ω, the solution of the integral equation
∂
Z
πσ(x) + σ(y) ln(|x − y|) dSy = −Φ.
∂Ω ∂ν(y)
1.6. EXERCISES 45
Find the density σ if Ω is a disk BR (0).
Hint: Show that

∂ 1
ln(|x − y|) =
∂ν(y) 2R
if x, y ∈ ∂BR (0). This formula is a consequence of
∂ 1
ln(|x − y|) = (y − x) · ν(y)
∂ν(y) |y − x|2
1
= cos β,
|y − x|
see Figure 1.11 for notations.
ν(y)
β
y
Figure 1.11: Notations to the exercise
8. Show that
C α [a, b] := {u ∈ C[a, b] : ||u||α < ∞},
where −∞ < a < b < ∞ and 0 < α ≤ 1, defines a Banach space,
where
|u(x) − u(y)|
||u||α := max |u(x)| + sup .
x∈[a,b] x,y∈[a,b],x6=y |x − y|α
9. Show that C ∞ [a, b] is not dense in C α [a, b], i. e., there is a u ∈ C α [a, b]
such that no sequence un ∈ C ∞ [a, b] exists such that ||un − u||α tends
to zero.
√ √
Hint: Consider u = x and [a, b] = [0, 1]. We have x ∈ C 1/2 [0, 1].
Assume
|un (x) − u(x) − (un (y) − u(y))|
sup →0
x,y∈[a,b],x6=y |x − y|α
if n → ∞, where α = 1/2. Then

un0 (x) − un0 (0)
√ − 1 ≤ ǫ
x
for a given ǫ > 0 and an integer n0 = n0 (ǫ).
10. Show that C0∞ (a, b) is not dense in C α [a, b].

Hint: Consider (a, b) = (−1, 1) and
√
x : 0≤x≤1
u(x) = .
0 : −1 ≤ x ≤ 0
Chapter 2
Perron’s method
Perron’s method is a maximum principle based existence theory for second

order linear or quasilinear elliptic equations. In this chapter we consider the
Dirichlet problem for the Laplace equation △u = 0 in Ω and u = Φ on ∂Ω,
where Ω ⊂ Rn is bounded and connected, and Φ is a given function defined
on ∂Ω.
In contrast to many other existence theories the Perron method provides
results under rather weak assumptions on the boundary ∂Ω since the prob-
lem of existence is separated from the question of the boundary behavior.
2.1 A maximum principle

We know that a harmonic function u must be a constant if u achieves its
supremum or infimum in a connected domain. This result is a consequence
of the mean value formula for harmonic functions, see [17], Chapter 7, for
example. Fortunately, there is a related principle for functions which satisfy
△u ≥ 0 or △u ≤ 0 throughout in Ω.
Lemma 2.1.1 (Mean value theorems). Suppose that u ∈ C 2 (Ω) satisfies

△u = 0, △u ≥ 0, △u ≤ 0 in Ω, resp. Then for any ball B = BR (x) ⊂⊂ Ω
1
Z
u(x) = (≤, ≥) u(y) dSy (2.1)
|∂B| ∂B
1
Z
u(x) = (≤, ≥) u(y) dy. (2.2)
|B| B
47
48 CHAPTER 2. PERRON’S METHOD
Proof. See [10], Chapter 2, for example. Let ρ ∈ (0, R) and Bρ = Bρ (x),
then
∂u
Z Z
△u dy = dSy
Bρ ∂Bρ ∂ν(y)
= (≥, ≤) 0,
respectively. Here is ν(y) the exterior unit normal at y on ∂Bρ .

Set r = |x − y|, ω = (y − x)/r, then u(y) = u(x + rω). Thus
∂u
Z Z
dSy = uyi (x + ρω)ωi dSy
∂Bρ ∂ν(y) ∂Bρ

∂u(x + rω)
Z
= dSy
∂Bρ ∂r
r=ρ

∂u(x + rω)
Z
n−1
= ρ dω
∂B1 (0) ∂r
r=ρ
∂
Z
= ρn−1 u(x + ρω) dω
∂ρ ∂B1 (0)
!
∂
Z
= ρn−1 ρ1−n u(y) dSy
∂ρ ∂Bρ
= (≥, ≤) 0.
Consequently for any ρ ∈ (0, R)

!
∂
Z
ρ1−n u(y) dSy = (≥, ≤) 0.
∂ρ ∂Bρ
It follows
Z Z
ρ1−n u(y) dSy = (≤, ≥) R1−n u(y) dSy
∂Bρ ∂BR
or
1 1
Z Z
u(y) dSy = (≤, ≥) u(y) dSy .
|∂Bρ | ∂Bρ |∂BR | ∂BR
Letting ρ → 0, we obtain
1
Z
u(x) = (≤, ≥) u(y) dSy .
|∂BR | ∂BR
2.1. A MAXIMUM PRINCIPLE 49
Formula (2.2) follows from

Z
n−1
ρ ωn u(x) = (≤, ≥) u(y) dSy ,
∂Bρ
where 0 < ρ ≤ R. We recall that ωn = |∂B1 (0)|. Integrating over (0, R), we
obtain
ωn R n
Z
u(x) = (≤, ≥) u(y) dy,
n BR
which is formula (2.2) since |BR | = Rn /(nωn ). ✷
As a consequence of Lemma 2.1.1 we get the following generalization of the

maximum principle for harmonic functions.
Theorem 2.1.1 (Strong maximum principle). Assume Ω ⊂ Rn is a con-

nected domain and u ∈ C 2 (Ω). Let △u ≥ 0 (△u ≤ 0) in Ω and suppose
there exists a point y ∈ Ω for which u(y) = supΩ u (u(y) = inf Ω u). Then u
is a constant.
Proof. Consider the case that △u ≥ 0 in Ω. Let x0 ∈ Ω such that
M := u(x0 ) = sup u(x).

x∈Ω
Set Ω1 = {x ∈ Ω : u(x) = M } and Ω2 = {x ∈ Ω : u(x) < M }. The set

Ω1 is not empty and the set Ω2 is open since u ∈ C(Ω). Consequently Ω2
is empty if we can show that Ω1 is an open set. Let y ∈ Ω1 , then there is a
ρ0 > 0 such that Bρ0 (y) ⊂ Ω and u(x) = M for all x ∈ Bρ0 (y). If not, then
there are ρ > 0, z ∈ Ω such that |z − y| = ρ, 0 < ρ < ρ0 and u(z) < M .
From Lemma 2.1.1 we have
1
Z
M ≤ u(x) dSx
ωn ρn−1 ∂Bρ (y)
M
Z
< u(x) dS = M,
ωn ρn−1 ∂Bρ (y)
which is a contradiction. ✷
Corollary. Assume Ω is connected and bounded and u ∈ C 2 (Ω) ∩ C(Ω)

satisfies △u ≥ 0 (△u ≤ 0) in Ω. Then u achieves its maximum (minimum)
on the boundary ∂Ω.
Corollary. Assume Ω is connected and bounded and v, w ∈ C 2 (Ω) ∩ C(Ω)

satisfy △v ≥ △w in Ω and v ≤ w on ∂Ω. Then v ≤ w in Ω.
Proof. Exercise.
2.2 Subharmonic, superharmonic functions

A function u ∈ C 2 (Ω) is called subharmonic (superharmonic) in a domain
Ω ∈ Rn if △u ≥ 0 (△u ≤ 0) in Ω. It turns out that we can define superhar-
monic and subharmonic functions if u merely is in C(Ω).
Definition. A function u ∈ C(Ω) is called subharmonic (superharmonic)

in Ω if for every ball B ⊂⊂ Ω and every function h harmonic in B, i. e.,
h ∈ C 2 (B) ∩ C(B) and △h = 0 in B, satisfying u ≤ h (u ≥ h) on ∂B we
have u ≤ h (u ≥ h) in B.
Corollary. A harmonic function in Ω is both a superharmonic and a sub-

harmonic function. In particular, constants are super- and subharmonic.
Remark. A function u in the class C 2 (Ω) is subharmonic (superharmonic)

if △(u − h) ≥ 0 (△(u − h) ≤ 0) in B for any harmonic function h in B such
that u ≤ h (u ≥ h) on ∂B.
Lemma 2.2.1 (Strong maximum principle for sub-, superharmonic func-

tions). Assume Ω is connected. If a subharmonic function u attends its
supremum in Ω, then u ≡ const. in Ω, and if a superharmonic function
attends its infimum in Ω, then u = const. in Ω.
Proof. Consider the case of a subharmonic function. Let x0 ∈ Ω and

M := u(x0 ) = sup u(x).
Ω
We will show that

Ω1 = {x ∈ Ω : u(x) = M }
is an open set. It is not empty since x0 ∈ Ω1 . Let x1 ∈ Ω1 , then Bρ0 (x1 ) ⊂
Ω1 if ρ0 > 0 is sufficiently small such that Bρ0 (x1 ) ⊂ Ω. If not, then there
is a ρ, 0 < ρ ≤ ρ0 and an x2 ∈ ∂Bρ (x1 ) such that u(x2 ) < M . Consider
a function h harmonic in B = Bρ (x1 ) and h = u on ∂B. Then, since h is
harmonic in B and u is subharmonic in Ω,
M ≥ max u = max h ≥ h(x) ≥ u(x),
∂B ∂B
2.2. SUBHARMONIC, SUPERHARMONIC FUNCTIONS 51
where x ∈ B. Consequently h(x1 ) = M since u(x1 ) = M . Thus h =

const. in B since the harmonic function h attends its maximum in B. Since
h(x) = u(x) on ∂Ω we have u(x) = M for all x ∈ ∂B, a contradiction to
u(x2 ) < M . ✷
The following lemma is a generalization of the comparison principle for

u, v ∈ C 2 (Ω) ∩ C(Ω) which says that △v ≤ △u in Ω and v ≥ u on ∂Ω
imply that either v > u throughout Ω or v ≡ u.
Lemma 2.2.2 (Comparison principle) Suppose that Ω is bounded and con-

nected. Let u ∈ C(Ω) be a subharmonic and v ∈ C(Ω) a superharmonic
function with u − v ≤ 01 on ∂Ω. Then either v > u throughout Ω or v ≡ u.
Proof. We will show that u − v ≡ const. in Ω if u − v attends a non-negative

supremum in Ω. Let x0 ∈ Ω such that
M := sup(u − v) = (u − v)(x0 ) ≥ 0.
Ω
Set Ω1 = {x ∈ Ω : (u − v)(x) = M }. This set is not empty since x0 ∈ Ω1 .

We will show that Ω1 is an open set. Let x1 ∈ Ω1 and consider a ball
Bρ0 (x1 ) ⊂⊂ Ω. Then Bρ0 (x1 ) ∈ Ω1 . If not, then there is a ball B = Bρ (x1 ),
0 < ρ ≤ ρ0 , and an x2 ∈ ∂B such that (u−v)(x2 ) < M . Let h1 , h2 harmonic
in B with h1 = u on ∂B and h2 = v on ∂B. Then, if x ∈ B,
M ≥ max(u − v) = max(h1 − h2 )
∂B ∂B
≥ h1 (x) − h2 (x) ≥ u(x) − v(x).
Set x = x1 , then by assumption u(x1 ) − v(x1 ) = M which implies that

the harmonic function h1 − h2 attends its maximum in B. Consequently
h1 −h2 = const. in B. Thus u(x)−v(x) = M on ∂B which is a contradiction
to u(x2 ) < M . We have seen that u − v ≡ M ≥ 0 in Ω. Finally the
assumption u − v ≤ 0 on ∂Ω implies u(x) ≡ v(x) in Ω. ✷
Let u be subharmonic in Ω, B ⊂⊂ Ω a ball and u harmonic in B such that

u = u on ∂B.
1
Here u − v ≤ 0 on ∂Ω means that
lim sup (u − v) ≤ 0
y→x, y∈Ω, x∈∂Ω
Definition. The function

u : x∈B
U (x) =
u(x) : x∈Ω\B
is called harmonic lifting of u in B.
Lemma 2.2.3. U is subharmonic in Ω.
Proof. Let B ′ ⊂⊂ Ω and h harmonic in B ′ with h ≥ U on ∂B ′ . We have to

show that h ≥ U in B ′ . For some of the following notations see Figure 2.1.
On C1 = ∂B ′ \ B we have h ≥ U ≡ u. On C3 = ∂B ′ ∩ B it is, according to
B’
B
h>U C2 U=u
C1
u=U
U=u
U>u
u=u=U
C3
h>U
u=u
Figure 2.1: Proof of Lemma 2.2.3
the definition of u, h ≥ U = u ≥ u. Combining these inequalities, we find

that h ≥ u on ∂B ′ which implies that h ≥ u in B ′ . Then
U ≤ h in B′ \ B (2.3)
since U ≡ u in B ′ \ B. It remains to show that also U ≤ h in B ′ ∩ B. On

∂(B ′ ∩ B) we have h ≥ U , see (2.3) and assumption h ≥ U on B ′ . Since
U ≡ u in B ∩ B ′ and h is harmonic in B ′ it follows that h ≥ U in B ∩ B ′ . ✷
Lemma 2.2.4. Let u1 , u2 , . . . , uN be subharmonic in Ω. Then
u(x) := max{u1 (x), . . . , uN (x)}

is also subharmonic in Ω, and if u1 , . . . , uN are superharmonic, then u(x) :=

min{u1 (x), . . . , uN (x)} is superharmonic in Ω.
Proof. Exercise.
Definition. Let Ω be bounded and φ a bounded function on ∂Ω. A subhar-

monic function u ∈ C(Ω) is called subfunction with respect to φ if u ≤ φ on
∂Ω, and a superharmonic function u ∈ C(Ω) is called a superfunction with
respect to φ if u ≥ φ on ∂Ω.
Here u ≤ φ on ∂Ω means that
lim sup u(y) ≤ φ(x).

Lemma 2.2.5. Suppose u is a subfunction and u a superfunction with

respect to φ. Then u ≤ u in Ω.
Proof. Lemma 2.2.2 and
lim sup (u(y) − u(y)) = lim sup (u(y) − φ(x) + φ(x) − u(y))
y→x, y∈Ω, x∈∂Ω y→x, y∈Ω, x∈∂Ω
≤ lim sup (u(y) − φ(x))
+ lim sup (φ(x) − u(y))
≤ 0.
Remark. The set of subfunctions with respect to φ and the set of super-
functions with respect to φ are not empty since constants ≤ inf ∂Ω φ are
subfunctions and constants ≥ sup∂Ω φ are superfunctions.
Set
Sφ = {v ∈ C(Ω) subharmonic in Ω : v ≤ φ on ∂Ω}.
Theorem 2.2.1 (Perron, [19]). The function
u(x) := sup v(x)

v∈Sφ
is harmonic in Ω.
Proof. (i) We have in Ω that
inf φ ≤ u(x) ≤ sup φ.

∂Ω ∂Ω
To show this inequality, let v ∈ Sφ , then v(x) ≤ φ(x) ≤ sup∂Ω φ on ∂Ω.

Since the constant sup∂Ω φ is a superfunction with respect to φ and v is a
subfunction with respect to φ we obtain from Lemma 2.2.5 the inequality
v(x) ≤ sup∂Ω φ, x ∈ Ω. Consequently u(x) ≤ sup∂Ω φ, x ∈ Ω. The other
side of the above inequality follows since the constant inf ∂Ω φ is an element
of Sφ .
(ii) Let y ∈ Ω be fixed. Then there is a sequence vn ∈ Sφ with limn→∞ vn (y) =

u(y). Let B = BR (y) ⊂⊂ Ω, R sufficiently small, and let Vn be the harmonic
lifting of vn in B. Then
Vn ∈ Sφ , (2.4)
and
lim Vn (y) = u(y). (2.5)
n→∞
Proof of (2.4): That Vn is subharmonic is the assertion of Lemma 2.2.3.

Since Vn = vn on ∂B, we have Vn ≤ φ on ∂B.
Proof of (2.5): We have
vn (y) ≤ Vn (y)
since vn = Vn on ∂B, Vn is harmonic in B and vn is superharmonic. Then
u(y) ≤ lim inf Vn (y).

n→∞
On the other hand, since Vn ∈ Sφ , we have
Vn (y) ≤ sup v(y) = u(y),

v∈Sφ
which implies that

lim sup Vn (y) ≤ u(y).
n→∞
(iii) For every function h harmonic in B we have,
sup |Dα h| ≤ C(ρ, R, α, n) sup |h|, (2.6)

Bρ (y) BR (y)
where 0 < ρ < R and the constant C is finite. This inequality is a conse-
quence of Poisson’s formula for the solution of the Dirichlet problem in a
ball, see [13, 10, 17], for example. Consequently for each fixed ρ, 0 ≤ ρ ≤ R
there exists a subsequence Vnk which converges uniformly in Bρ (y) to a
harmonic function v. It follows that there is a subsequence of Vn , denoted
also by Vnk , which converges uniformly on compact subsets of BR (0) to a
function v harmonic in BR (0). We have
v(x) ≤ u(x), x ∈ BR (y) (2.7)
since Vnk (x) ≤ u(x) on BR (y), see the definition of u(x). At the center y it
is, see (2.5),
v(y) = u(y). (2.8)
(iv) Claim: v(x) = u(x), x ∈ B.

Proof: If not, then there is a z ∈ B such that v(z) < u(z). Then there exists
an u0 ∈ Sφ with v(z) < u0 (z). Set
wk (x) := max(u0 (x), Vnk (x)).
Let Wk be the harmonic lifting of wk in B. A subsequence of Wk converges

uniformly on each compact subset of B to a function w harmonic in B such
that
v(x) ≤ w(x) ≤ u(x), x ∈ B = BR (y). (2.9)
These inequalities follow since wnk (x) ≤ Wnk (x), wnk (x) ≥ Vnk (x) and
Wnk (x) ≤ u(x), where x ∈ B.
Combining equation (2.8) and inequalities (2.9), we obtain
v(y) = w(y) = u(y). (2.10)
Thus, the harmonic function v − w is less or equal zero in B and zero in the
interior point y ∈ B. The strong maximum principle implies that
v(x) = w(x), x ∈ B. (2.11)
According to the assumption we have for a z ∈ B
v(z) < u0 (z).
On the other hand, see the definition of wn and Wn , if x ∈ B then
u0 (x) ≤ wnk (x) ≤ Wnk (x),
which implies that

u0 (x) ≤ w(x), x ∈ B.
Summarizing, we have for the particular z under consideration the inequal-

ities
v(z) < u0 (z) ≤ w(z),
which is a contradiction to (2.11). ✷
2.3 Boundary behavior

One of the advantages of Perron’s method is that the boundary behavior of
solutions is separated from the existence problem.
Definition. A C(Ω)-function w = wξ is called a barrier at ξ ∈ ∂Ω relative

to Ω if
(i) w is superharmonic in Ω,
(ii) w > 0 in Ω \ {ξ} and w(ξ) = 0.
w is called a local barrier at ξ ∈ ∂Ω if there is a neighborhood N of ξ such

that w satisfies (i) and (ii) in Ω ∩ N instead in Ω.
Let w be a local barrier at ξ ∈ ∂Ω, then we can define a barrier at ξ ∈ ∂Ω

relative to Ω as follows. Let B = BR (ξ), R > 0 sufficiently small such that
B ⊂⊂ N , see Figure 2.2. Set
ξ
.
N
B
Figure 2.2: Definition of a local barrier
m = inf w
N \B
2.3. BOUNDARY BEHAVIOR 57
. We have m > 0, see assumption (ii) in the above definition.
Lemma 2.2.6. The function

min(m, w(x)) : x∈Ω∩B
w0 (x) =
m : x∈Ω\B
is a barrier at ξ relative to ∂Ω.
Proof. The property w0 ∈ C(Ω) follows since w0 = m on Γ = ∂B ∩ Ω. If

not, then there is an x0 ∈ Γ with w(x0 ) < m, which is a contradiction to
the definition of m. Now we will show that w0 is superharmonic in Ω. This
follows since
w1 : x ∈ Ω ∩ B
w0 (x) = ,
w2 : x ∈ Ω \ B
where w1 (x) = min(m, w(x)), w2 = m and w1 , w2 are superharmonic in
Ω ∩ B and Ω \ B, respectively, and since w1 = w2 on Γ. To show this,
consider a ball B ′ ⊂⊂ Ω located as shown in Figure 2.3. Let h be harmonic
w0 =m
ξ
. B´
Figure 2.3: A local barrier defines a barrier
in B ′ with h ≤ w0 on ∂B ′ . We have to show that h ≤ w0 in B ′ . Since

w0 ≤ m on ∂1 B ′ = ∂B ′ ∩ B and w0 = m on ∂2 B ′ = ∂B ′ \ ∂1 B ′ , we have
w0 ≤ m on ∂B ′ . Thus, the assumption h ≤ w0 on ∂B ′ implies that h ≤ m
on ∂B ′ . In particular, h ≤ w0 in B ′ \ B since w0 = m on B ′ \ B. Finally we
have h ≤ w0 in B ∩ B ′ since h ≤ w0 on B ′ ∩ ∂B and on B ∩ ∂B ′ . We recall
that w0 = w1 in Ω ∩ B and w1 is superharmonic in Ω ∩ B, see Lemma 2.2.4.
✷
Definition. A boundary point is said to be regular if there exists a barrier

at that point.
Lemma 2.2.7. Let u be a harmonic function defined in Ω by the Perron

method with boundary data φ. If ξ is a regular point af ∂Ω and if φ is
continuous at ξ, then
lim u(x) = φ(ξ).
x→ξ,x∈Ω
Proof. Fix ǫ > 0. Then there is a δ = δ(ǫ) > 0 such that |φ(x) − φ(ξ)| < ǫ
for all x ∈ ∂Ω satisfying |x − ξ| < δ. Set M = sup∂Ω |φ|. Let w be a barrier
at ξ. Then there is a k = k(ǫ) > 0 such that kw(x) > 2M if |x − ξ| ≥ δ.
Step 1. We will show that φ(ξ) + ǫ + kw(x) is a superfunction relative to φ.

We recall that w ∈ C(Ω), w is superharmonic in Ω, w > 0 in Ω \ {ξ} and
w(ξ) = 0. Then
φ(ξ) + ǫ + kw(x) ≥ φ(x)
on ∂Ω, since for x ∈ ∂Ω with |x − ξ| ≥ δ we have
φ(ξ) + ǫ + kw(x) ≥ φ(ξ) + ǫ + 2M

≥ φ(x),
and for x ∈ ∂Ω with |x − ξ| < δ it is
φ(x) − φ(ξ)| ≤ ǫ
since |φ(x) − φ(ξ)| ≤ ǫ if x ∈ ∂Ω ∩ Bδ (ξ) and kw(x) ≥ 0.
Step 2. Since
u(x) = sup v(x)
v∈Sφ
and φ(ξ) − ǫ − kw(x) is a subfunction relative to φ, we have
φ(ξ) − ǫ − kw(x) ≤ u(x)
in Ω. The function φ(ξ) + ǫ + kw(x) is a superfunction relative to φ, see

Step 1. Consequently
v(x) ≤ φ(ξ) + ǫ + kw(x)
for all v ∈ Sφ , see Lemma 2.2.5. This inequality implies that
u(x) ≤ φ(ξ) + ǫ + kw(x).

Summarizing, we get
|u(x) − φ(ξ)| ≤ ǫ + kw(x).
Since limx→ξ,x∈Ω w(x) = 0, we obtain finally
lim u(x) = φ(ξ).

x→ξ,x∈Ω
Theorem 2.2.2. Let Ω ⊂ Rn be bounded and connected. Then the Dirichlet

problem △u = 0 in Ω, u = φ on ∂Ω, where φ ∈ C(∂Ω) is given, is solvable
if and only if the boundary points are all regular.
Proof. (i) Assume all points of ∂Ω are regular points. Then the assertion
follows from previous Lemma 2.2.7.
(ii) Assume the Dirichlet problem is solvable for all continuous φ ∈ C(∂Ω).
Set φ(x) := |x − ξ| and consider the Dirichlet problem with the boundary
condition u(x) = φ(x) on ∂Ω. Let uξ (x) be the solution, then uξ (x) is a
barrier at ξ. Consequently all boundary points are regular.
✷
2.3.1 Examples for local barriers

Slit domains in R2
Let Ω ⊂ R2 with a slit along the negative x−axis at x = 0 as indicated in
Figure 2.4. Let Ln z := ln |z| + iφ, −π < φ ≤ π. Then
x
2
x1
Figure 2.4: A slit domain


1 ln r
w := −Re =−
Lnz ln2 r + φ2
is a local barrier at ξ = 0. Here w(0) is defined as the limit w(0) :=

limz→0,z∈Ω w(z). We have w(0) = 0 and w(x) > 0 in Ω ∩ BR (0). For higher
dimensions there are counterexamples. One of them was given by Lebesgue,
see [4], Part II, p. 272, for example, which shows that sufficiently sharp
cusps are not regular at the tip of the cusp, see Figure 2.5.
Figure 2.5: A cusp boundary point
Exterior sphere condition

We say that Ω satisfies the exterior sphere condition at ξ ∈ ∂Ω if there is a
sphere BR (y) ⊂ Rn \ Ω such that BR (y) ∩ Ω = {ξ}, see Figure 2.6. Let Ω
Figure 2.6: Exterior sphere condition
satisfies the exterior sphere condition at ξ ∈ ∂Ω, then

( 2−n
R − |x − y|2−n : n≥3
w(x) =
|x−y|
ln R : n=2
is a local barrier at ξ.
Exterior cone condition

We say that Ω satisfies an exterior cone condition at ξ ∈ ∂Ω if there is
a finite circular cone C with the vertex at ξ such that K ∩ Ω = {ξ}, see
Figure 2.7. Let ξ be the origin and assume the exterior cone property is
Figure 2.7: Exterior cone condition
satisfied at ξ ∈ ∂Ω. Then we can find a positive constant λ and a positive

function f (θ), where θ is the polar angle, such that
w = rλ f (θ),
r = |x|, is a local barrier at ξ.
Two-dimensional domains. Here we find λ and f (θ) as follows. Let
C ⊂ {(r, θ) : r > 0, −α < θ < α},
see Figure 2.8. Introducing polar coordinates (r, θ), where
x1 = r sin θ, x2 = r cos θ,
we have
w(x) = W (r, θ) := w(r cos θ, r sin θ)

1 ∂ 1
△w = (rWr ) + 2 Wθθ .
r ∂r r
Consider the ansatz
W (r, θ) = rλ cos(µθ),
where λ and µ are positive constants, then
△w = rλ−2 (λ2 − µ2 ) cos(µθ).

x
2
x
1
Figure 2.8: Local barrier, R2
Consequently △w ≤ 0 on C if λ ≤ µ and |µθ| ≤ π/2 for all θ satisfying

α < θ < 2π − α (0 < α < π/2).
Then w > 0 if r > 0 and α < θ < 2π − α. Then w is a local barrier at
the origin if we choose λ = µ with a sufficiently small positive µ.
Higher dimensional case. Let M ⊂ ∂B1 (0) be the manifold as indicated in

Figure 2.9. Consider the eigenvalue problem
−△′ v = ν 2 v in M
v = 0 on ∂M,
where △′ is the Laplace-Beltrami differential operator on the unit sphere.
We recall that in the two-dimensional case
∂2
△′ = 2
∂θ
and in the three-dimensional case
1 ∂2

′ 1 ∂ ∂
△ = sin θ + .
sin θ ∂θ ∂θ sin2 θ ∂φ2
For the definition of the Laplace-Beltrami operator see for example [2]
and for n-dimensional polar coordinates see [7], Part III, pp. 395, for exam-
ple.
2.4. GENERALIZATIONS 63
xn
θ0
M
Figure 2.9: Local barrier, Rn
Let ν1 be the first eigenvalue of the above eigenvalue problem. It is known

that ν1 is positive, a simple eigenvalue and the associated eigenfunction v1
has no zero in M. Thus, we can assume that v1 > 0 in M. Set
W = Arκ v1 ≡ w(x)
where A and κ are positive constants. Then w > 0 in Rn \ C and
△w = Arκ−2 v1 (κ(κ − 1) + (n − 1)κ − α1 ) ,
where s 2
n−2 n−2
α1 = − + + ν12 .
2 2
Consequently we have △w ≤ 0 in Rn \C, provided κ > 0 is sufficiently small.
2.4 Generalizations
Perron’s method can be applied to the Dirichlet problem for a more general
class of linear elliptic equations of second order, see for example [10], pp. 102.
The previous discussion in the case of the Laplace equation shows that we
need a strong maximum principle, the existence of solutions of the Dirichlet
problem on a ball with continuous boundary data and some estimates for
the derivatives. Then we are able to prove the existence of a solution of
the equation in the given domain. The problem of the boundary behavior
requires an additional discussion.
2.5. EXERCISES 65
2.5 Exercises
1. Let Ω ⊂ Rn be a connected domain. Consider the eigenvalue problem
−△u = λu in Ω
u = 0 on ∂Ω.
Suppose λ0 ≥ 0 is an eigenvalue and u0 ∈ C 2 (Ω) ∩ C(Ω) an associated

eigenfunction satisfying u0 (x) ≥ 0 in Ω. Show that u0 (x) > 0 in Ω.
2. Prove the second corollary to Theorem 2.1.1.
3. Prove Lemma 2.2.4.

Chapter 3
Maximum principles
Maximum principles provide powerful tools for linear and nonlinear elliptic
equations of second order. In these notes we consider linear equations.
3.1 Basic maximum principles

Set
n
X n
X
Mu = aij (x)Dij u + bi (x)Di u
i,j=1 i=1
Lu = M u + c(x)u,
where ai,j , bi and c are real and defined on a simply connected domain
Ω ⊂ Rn . We assume aij = aji . Let λ(x) be the minimum of the eigenvalues
of the symmetric matrix defined by the coefficients aij and let Λ(x) be the
maximum of these eigenvalues.
Definition. L is called elliptic in Ω if λ(x) > 0 in Ω. L is said to be strictly

elliptic in Ω if λ(x) ≥ λ0 > 0 in Ω, where λ0 is a constant. An elliptic L is
called uniformly elliptic if Λ/λ is bounded in Ω.
In the following we suppose that L is at least elliptic, and for each i

|bi (x)|
sup < ∞. (3.1)
x∈Ω λ(x)
Theorem 3.1.1 (Weak maximum principle). Let L be elliptic in the bounded

domain Ω. Assume a function u ∈ C 2 (Ω)∩C(Ω) satisfies M u ≥ 0 (M u ≤ 0)
in Ω. Then the supremum (infimum) of u on Ω is achieved on ∂Ω.
67
68 CHAPTER 3. MAXIMUM PRINCIPLES
Proof. Assume initially that M u > 0 in Ω. Then u cannot achieves an

interior maximum since ∇u(x0 ) = 0 at this point where u achieves its maxi-
mum, and since the matrix D2 u(x0 ) = [Dij u(x0 )] is non-positive (necessary
condition of second order). It follows that, see an exercise in [17], for exam-
ple,
Xn
M u(x0 ) = aij (x0 )Dij u(x0 ) ≤ 0
i,j=1
since the matrix [aij (x0 )] is non-negative (even positive) by assumption.

This inequality is a contradiction to our assumption.
For a positive sufficiently large constant γ we calculate
M eγx1 = (γ 2 a11 + γb1 )eγx1
≥ λ(γ 2 − γc1 )eγx1 > 0.
We recall that a11 ≥ λ and |b1 |/λ ≤ c1 , where c1 is a positive constant, see
assumption (3.1). Consequently for any ǫ > 0 we have in Ω
M (u + ǫeγx1 ) > 0.
Using the above result, we conclude that
sup (u + ǫeγx1 ) = sup (u + ǫeγx1 ) .
Ω ∂Ω
Letting ǫ → 0, we obtain
sup u = sup u.
Ω ∂Ω
✷
The next theorem is the strong maximum principle. It follows from the
boundary point lemma due to E. Hopf [11]. The proof of this lemma needs
the previous weak maximum principle. The strong maximum principle is
the essential tool to show existence of a solution of the Dirichlet problem
via Perron’s method.
Lemma 3.1.1 (E. Hopf, 1952). Let L be uniformly elliptic. Assume u ∈

C 2 (Ω) satisfies M u ≥ 0 in Ω. Let x0 ∈ ∂Ω and suppose that
(i) u is continuous at x0 ,
(ii) u(x0 ) > u(x) for all x ∈ Ω ∩ Ba (x0 ) for an a > 0,
(iii) ∂Ω satisfies the interior sphere condition at x0 .
Then the outer normal derivative of u at x0 , if it exists, satisfies
∂u
(x0 ) > 0.
∂ν
3.1. BASIC MAXIMUM PRINCIPLES 69
Proof. Let B = BR (y) be the ball related to the interior sphere condition,
see Figure 3.1. Consider the function
ν
. x0
ρ .
y
B (y)
R .x
Figure 3.1: Proof of Hopf’s boundary point lemma
2 2
v(x) = e−αr − e−αR ,
where r = |x − y| > ρ and α is a positive constant which we will determine

later. A calculation leads to
n
−αr 2
X
2
Mv = e 4α aij (xi − yi )(xj − yj )
i,j=1
n n
X X
−2α aii + bi (xi − yi )
i=1 i=1
n
−αr 2
X
≥ e 4α2 λ(x)r2 − 2α aii + |b|r ,
i=1
where b = (b1 , . . . , bn ). Since by assumption aii /λ and |b|/λ are bounded,

we may choose α large enough such that M v ≥ 0 in the annular domain
A := BR (y) \ Bρ (y). Since u(x) − u(x0 ) < 0 on ∂Bρ (y) there is a constant
ǫ > 0 such that u(x) − u(x0 ) + ǫv(x) ≤ 0 on ∂Bρ (y). This inequality is
also satisfied on ∂BR (y) by assumption

on u and since v = 0 on ∂BR (y).
We have M u(x) − u(x0 ) + ǫv(x) = M u + ǫM v. Then the weak maximum
principle implies that u − u(x0 ) + ǫv ≤ 0 in A. Thus

u(x0 ) − u(x) ≥ −ǫ v(x0 ) − v(x) ,
where x ∈ A and on the line defined by x0 and ν. Set

2 2
V (r) = e−αr − e−αR .
It follows that
∂u
(x0 ) ≥ −ǫV ′ (R),
∂ν
provided the normal derivative exists. ✷
Remark. If the normal derivative does not exist, then
u(x0 ) − u(x)
lim inf > 0,
x→x0 |x0 − x|
where the angle between x0 −x and the exterior normal ν is less then (π/2)−δ
for a fixed δ > 0.
Corollary. Suppose that ∂Ω satisfies the interior sphere condition at x0 ∈

∂Ω, u ∈ C 2 (Ω) ∩ C(Ω) satisfies Lu ≥ 0 and u(x0 ) > u(x) in Ω ∩ U , where U
is a neighborhood of x0 . If additionally c ≤ 0 in Ω ∩ U and u(x0 ) > 0 then
∂u/∂ν(x0 ) > 0, provided the normal derivative exists.
Proof. M u = Lu − cu ≥ 0 in Ω ∩ V, where V is a neighborhood of x0 . ✷
In generalization to the strong maximum principle for △ we have
Theorem 3.1.2 (Strong maximum principle). Let L be uniformly elliptic.

Assume u ∈ C 2 (Ω)∩C(Ω) satisfies M u ≥ 0 (M u ≤ 0) in a connected domain
Ω, not necessarily bounded. Then if u achieves its supremum (infimum) in
the interior of Ω, u is a constant.
Proof. Consider the case of a maximum. Assume u is not constant and

achieves its maximum m in the interior of Ω. Set Ω1 = {x ∈ Ω : u(x) = m}
and Ω2 = {x ∈ Ω : u(x) < m}. By assumption Ω1 is not empty. We will
show that Ω1 is open. Then Ω2 = ∅ since we suppose that Ω is connected.
Let x1 ∈ Ω1 . Consider a ball B = B2ρ0 (x1 ) ⊂⊂ Ω. If Ω1 is not open, then
2ρ
0
ρ x1
0 .
2
3 . .
x
x
Figure 3.2: Proof of the strong maximum principle
there is an x2 ∈ Bρ0 (x1 ) such that u(x2 ) < m. Consequently there is a ball
Bρ (x2 ), where 0 < ρ ≤ |x2 − x1 |, and u(x) < m in Bρ (x2 ) and there is an
x3 ∈ ∂Bρ (x2 ) such that u(x3 ) = m. See Figure 3.2 for notations. Hopf’s
lemma says that ∂u/∂ν > 0 at x3 , where ν is the exterior normal derivative
on x3 ∈ ∂Bρ (x2 ) at u(x3 ) which is a contradiction to the fact that u attends
an interior maximum at x3 . ✷
In many cases the assumption c ≡ 0 in Ω is not satisfied. If c(x) ≤ 0 in

Ω, then we have the following corollary to the previous theorem. If c(x) is
positive on a subset of Ω, then the situation is more complicated. In this
case one has to study an associated eigenvalue problem.
Corollary. Let Ω be a connected domain, not necessarily bounded. Suppose

L is uniformly elliptic and c(x) ≤ 0 in Ω. Assume u ∈ C 2 (Ω)∩C(Ω) satisfies
Lu ≥ 0 (Lu ≤ 0). If u achieves its positive supremum (negative infimum)
in the interior of Ω then u is a constant.
Proof. Consider the case of a maximum. Set m = supΩ u(x) and Ω1 = {x ∈

Ω : u(x) = m}, Ω2 = {x ∈ Ω : u(x) < m}. By assumption Ω1 is not empty.
We show that Ω1 is an open set. Let x1 ∈ Ω1 . Then there is ball Bρ (x1 ) ⊂ Ω
where u is non-negative. Thus
M u ≡ Lu − c(x)u ≥ 0
in Bρ (x1 ). The above strong maximum principle (Theorem 3.1.2) says that
u(x) = m for all x ∈ Bρ (x1 ).
✷
From this corollary follows a result which is important for many appli-
cations:
Theorem 3.1.3 (Comparison principle). Let Ω be a bounded and connected

domain. Suppose that L is uniformly elliptic and c(x) ≤ 0 in Ω. Assume
u, v ∈ C 2 (Ω) ∩ C(Ω) and satisfy Lu ≥ Lv in Ω and u ≤ v on ∂Ω. Then
u ≤ v in Ω.
Proof. Set w = u − v. Then Lw ≥ 0 in Ω and w ≤ 0 on ∂Ω. From the above

corollary we see that w can not achieve a positive maximum in Ω. ✷
3.1.1 Directional derivative boundary value problem

As an application of the previous corollary we consider a generalization of
the Neumann problem. Let Ω ⊂ Rn be a bounded and connected domain,
and assume ∂Ω is sufficiently smooth. Consider
Lu = f, in Ω (3.2)
∂u
= φ on ∂Ω, (3.3)
∂α
where f, φ are given and sufficiently regular, and the direction α is not
tangential on ∂Ω at each point of ∂Ω, see Figure 3.3.
ν
t
x α
0
Figure 3.3: Directional derivative boundary value problem

Proposition 3.1.1. Suppose that c ≤ 0 in Ω and let u1 , u2 ∈ C 2 (Ω)∩ C 1 (Ω)

are solutions of (3.2), (3.3). Then u1 − u2 ≡ const. in Ω.
Proof. Set u = u1 − u2 , then Lu = 0 in Ω and ∂u/∂α = 0 on ∂Ω. Suppose

that u is not constant, then we can assume maxΩ u > 0. This maximum
is achieved at x0 ∈ ∂Ω and u(x0 ) > u(x) for all x ∈ Ω, see the corollary
to the strong maximum principle. For the tangential derivative we have
(∂u/∂t)(x0 ) = 0, and by assumption implies that (∂u/∂α)(x0 ) = 0. Let
ν = a t + b α, then
(∂u/∂ν)(x0 ) = a(x0 )(∂u/∂t)(x0 ) + b(x0 )(∂u/∂α)(x0 )

= 0,
which is a contradiction to the corollary to the Hopf boundary point lemma

which says that ∂u/∂ν > 0 at x0 . ✷
3.1.2 Behavior near a corner

P2
Set Lu = i,j=1 aij (x)uxi xj and consider the Dirichlet problem
Lu = f in Ω (3.4)
u = 0 on ∂Ω, (3.5)
where f is given. Suppose that the boundary of Ω ⊂ R2 has a corner.

Without loss of generality we suppose that the corner is the origin. Set
Ωρ = Ω ∩ Bρ (0). Then we assume that there is a ρ > 0 such that in Ωρ
we have aij = aji , L is uniformly elliptic, aij ∈ C α (Ωρ ). An appropriate
rotation with center at the origin and a stretching of the axis transforms
(3.4), (3.5) into a Dirichlet problem where aij (0, 0) = δij . Here we denote
the transformed coefficients aij (C −1 y) and the transformed right hand side
f (C −1 y) by aij and f , resp., and y by x. The new domain is denoted by Ω
again. The new interior angle ω can be calculated from the original interior
angle γ and the original coefficients [aij (0, 0)], see an exercise. After this
mapping we arrived at the problem
2
X
Lu = △u + (aij (x) − δij ) uxi xj = f in Ωρ (3.6)
i,j=1
u = 0 on ∂Ω ∩ Ωρ , (3.7)
where
|aij (x) − δij | ≤ c|x|α .
Suppose that Ωρ is contained in a domain, see Figure 3.4, defined by 0 <

r < ρ and θ1 (ρ) ≤ θ ≤ θ2 (ρ), where r = (x21 + x22 )1/2 . Set
ω = ω(ρ) = θ2 (ρ) − θ1 (ρ)
and consider the function

x2
Ωρ
ρ
π−θ (ρ)
2 θ1 (ρ)
x1
Figure 3.4: Corner domain
π
v(θ) = sin − h(ǫ) (θ − θ1 (ρ) + ǫ) ,
ω
where 0 < ǫ ≤ ǫ0 , ǫ0 sufficiently small, and h(ǫ) = 2πǫ/ω 2 . There is an
ǫ0 > 0 and a positive constant c, independent on ǫ such that v(θ) ≥ cǫ for
all 0 < ǫ ≤ ǫ0 and θ1 ≤ θ ≤ θ2 (exercise). Set
π
κ= − h(ǫ)
ω
and consider the function
W (r, θ) = Arκ−η v(θ)

= w(x1 , x2 ),
where A and η are constants, 0 < η < κ. In polar coordinates we have
1 ∂2W

1 ∂ ∂W
△w = r + 2 .
r ∂r ∂r r ∂θ2
Then
△w = Arκ−η−2 (κ − η)2 − κ2 v

and
|wxi xj | ≤ c|A|rκ−η−2 .
The constant c, here and in the following formulas, are independent on r
and θ. Then
Lw = Arκ−η−2 (κ − η)2 − κ2 v + O Arκ−η−2+α .

Suppose that the constant A is positive, then from the above considerations
it follows that for given δ > 0, 0 < δ ≤ δ0 , δ0 sufficiently small, there are
positive constants c(δ) and ρ = ρ(δ) such that
Lw ≤ −Ac(δ)r(π/ω)−δ−2
in Ωρ .
Proposition 3.1.2. Assume the right hand side f of (3.4) satisfies
|f | ≤ cr(π/ω)−2−δ+τ
in Ωρ0 , ρ0 > 0, for a τ > 0. Assume u ∈ C 2 (Ωρ0 ) and supΩρ0 |u(x)| < ∞.
Then for given (small) ǫ > 0 there exists positive constants c(ǫ) and ρ(ǫ) > 0
such that
|u| ≤ c(ǫ)|x|(π/ω)−ǫ
in Ωρ(ǫ) .
Proof. We have Lw ≤ Lu in Ωρ and w ≥ u on ∂Ωρ , provided ρ > 0 is

sufficiently small. ✷
Remark. The additional assumption that u remains bounded up to the

corner is essential for the previous proposition since there exists also solu-
tions which are unbounded near the corner. An example is the boundary
value problem △u = 0 in Ωα , u = 0 on ∂Ωα , where Ωα is the sector defined
by r > 0 and 0 < θ < α, where 0 < α < 2π. Solutions are given by
u(x) = rπk/α sin ((π/α)kθ) ,
where k ∈ {±1, ±2, . . .}. For a class of quasilinear non-uniformly bound-

ary value problems the behavior of the solution near the corner does not
require such an additional assumption near the boundary, see [8, 16, 3, 18].
The behavior follows from the problem itself. The reason for this striking
difference is that essential information is lost through the linearization.
Remark. Asymptotic expansions near a corner yield more precise behavior

near the corner, see [15] for a class of uniformly quasilinear elliptic Dirichlet
problems. In general, the expansion, depends on the solution considered, in
contrast to some nonuniform problems, see [16].
3.1.3 An a priori estimate

Consider for given bounded functions f , Φ defined on Ω and on ∂Ω, respec-
tively, the Dirichlet problem
n
X n
X
Lu = aij (x)Dij u + bi (x)uxi + c(x) = f (x) in Ω (3.8)
i,j=1 i=1
u = Φ on ∂Ωy (3.9)
where ai,j , bi and c are real and defined on a simply connected and bounded
domain Ω ⊂ Rn . We assume aij = aji , L is strictly elliptic and
sup |bi (x)| < ∞
Ω
for every i = 1, . . . , n. Let K be a bound of b1 and set
1
α= K + (K 2 + 4λ0 )1/2 , (3.10)
λ0
where the positive constant λ0 is a lower bound of the minimum of the
eigenvalues of the matrix [aij (x)]. Set d = diam Ω.
Proposition 3.1.3. Suppose that c(x) ≤ 0 in Ω and let u ∈ C 2 (Ω) ∩ C(Ω)

be a solution of (3.8), (3.9). Then

sup |u| ≤ sup |Φ| + eαd − 1 sup |f |.
Ω ∂Ω Ω
Proof. Suppose that Ω is contained in the strip defined by 0 ≤ x1 ≤ d.

Consider the function
g(x) = eαd − eαx1 ,
where α is positive constant which will be determined later. We have g(x) ≥
0 if x ∈ Ω. We get
Lg = −(a11 α2 + b1 α)eαx1 + cg
≤ (λ0 α2 + Kα)eαx1
≤ −eαx1
≤ −1,
3.2. A DISCRETE MAXIMUM PRINCIPLE 77
provided that α is large enough. We can choose α given by (3.10). Set
h = sup |Φ| + g(x) sup |f |.

∂Ω Ω
Then
Lh = Lg sup |f | + c sup |Φ|

Ω ∂Ω
≤ − sup |f |.
Ω
We recall that c ≤ 0 on Ω. Summarizing, we have
Lh ≤ − sup |f | in Ω
Ω
h ≥ sup |Φ| on ∂Ω.
Ω
Set v = u − h, then
Lv = f − Lh ≥ f + sup |f | ≥ 0 in Ω
Ω
v = Φ − sup |Φ| ≤ 0 on ∂Ω.
∂Ω
The comparison principle says that v ≤ 0 in Ω. The same argument leads

to the inequality u ≥ −h in Ω if we set v = −u − h. Then
Lv = −f − Lh ≥ −f + sup |f | ≥ 0 in Ω
Ω
v = −Φ − h ≤ −Φ − sup |Φ| ≤ 0 on ∂Ω.
∂Ω
3.2 A discrete maximum principle

To simplify the presentation we consider here a subclass of elliptic boundary
value problems in a domain Ω ∈ R2 , see [12], pp. 458, for example. Set
M u = △u + b1 (x, y)ux + b2 (x, y)uy .
Suppose that there is a constant K such that
sup (|b1 (x, y)| + |b2 (x, y)|) ≤ K. (3.11)

Ω
Let
Lu = M u + c(x, y)u,
where c(x, y) is defined on Ω. Then we consider for given f defined on Ω,

and Φ defined on ∂Ω, the Dirichlet problem
Lu = f in Ω (3.12)
u = Φ on ∂Ω. (3.13)
Let h > 0 be a (small) constant. We define an associated difference operator

to M u by
u(x + h, y) + u(x − h, y) + u(x, y + h) + u(x, y − h) − 4u(x, y)

Mh u =
h2
u(x + h, y) − u(x − h, y) u(x, y + h) − u(x, y − h)

+b1 (x, y) + b2 (x, y) .
2h 2h
We have limh→0 Mh u = M u on every compact subdomain of Ω, provided
u ∈ C 2 (Ω).
Set P0 = (x, y) and assume P0 ∈ Ω. The four points P01 = (x + h, y),
P02 = (x, y + h), P03 = (x − h, y), P04 = (x, y − h) are called h-neighborhood
of P0 or a star around P0 . Consider the intersection of Ω with an h-net Nh
of R2 defined by
Nh (x, y) = {(x + lh, y + kh) ∈ R2 : k, l = 0, ±1, ±2, . . .},
where (x, y) ∈ R2 is given. Let Ωh be the set of the points Pi of Ω ∩ Nh such

that each point of the star associated to Pi are contained in Ω. The set of
all star points not in Ωh is denoted by ∂Ωh , see Figure 3.5. Then we define
Ωh = Ωh ∪ ∂Ωh .
Suppose that Ωh is connected in the sense that for given P, R ∈ Ωh there
exists points Q1 , Q2 , . . . , Qs ∈ Ωh , where Q1 = P , and Qs = R, such that
Qj is a point of the h-neighborhood of Qj−1 , j = 1, . . . , s + 1. We assume
that h > 0 is sufficiently small such that
hK < 1. (3.14)
Theorem 3.2.1. Let u be defined on Ωh . Suppose that Mh u ≥ 0 in Ωh and

u attends its maximum in Ωh . Then u is constant on Ωh .
Figure 3.5: Definition of Ωh and ∂Ωh
Proof. Let Pi ∈ Ωh . From the definition of Mh u and the assumption Mh u ≥

0 we obtain that
X4
u(Pi ) ≤ λij u(Pij ),
j=1
where Pij are the points of the h-neighborhood of Pi and

1 hb1 (Pi ) 1 hb2 (Pi )
λi1 = 1+ , λi2 = 1+
4 2 4 2

1 hb1 (Pi ) 1 hb2 (Pi )
λi3 = 1− , λi4 = 1− .
4 2 4 2
P4
We have j=1 λij = 1 and, see (3.14), λij > 0. Assume
m := max u(x) = u(Pi ).

Ωh
Then u = m in all points of the h-neighborhood of Pi . Since Ωh is connected,

the theorem is shown. ✷
Set Lh u = Mh u + c(x)u. The following corollary is the discrete version of

the corollary to the above Theorem 3.2.
Corollary. Suppose that c(x) ≤ 0 on Ωh and u defined on Ωh satisfies

Lh u ≥ 0 on Ωh . If u achieves its non-negative supremum m in Ωh then u
is constant on Ωh .
Proof. Set Ωh,1 = {x ∈ Ωh : u(x) = m} and Ωh,2 = {x ∈ Ωh : u(x) < m}.

By assumption Ωh,1 is not empty. The set Ωh,1 is open in the sense that if
Pi ∈ Ωh,1 then u(Pil ) = m for every point Pil , l = 1, 2, 3, 4, of the associated
star to Pi . This follows from the inequality
3
X h2
u(Pi ) ≤ λij u(Pij ) + c(Pi )u(Pi ).
4
j=1
Thus u = m on Ωh since Ωh is connected by assumption. ✷
From this corollary it follows
Theorem 3.2.2 (Comparison principle). Let Ωh be a bounded and connected

domain. Suppose that c(x) ≤ 0. Assume u, v are defined on Ωh and satisfy
Lh u ≥ Lh v in Ωh and u ≤ v on ∂Ωh . Then u ≤ v on Ω.
Proof. Set w = u − v. Then Lh w ≥ 0 in Ωh and w ≤ 0 on ∂Ωh . From the

above corollary we see that w can not achieve a nonnegative maximum in
Ω. ✷
Suppose that Ωh is bounded and connected. We consider the discrete Dirich-

let problem
Lh u = f in Ωh (3.15)
u = Φ on ∂Ωh , (3.16)
where f is defined in Ωh and Φ on ∂Ωh . Assume c ≤ 0 and h is sufficiently

small such that the inequality (3.14) is satisfied.
Corollary. There exists a unique solution of the discrete Dirichlet prob-

lem (3.15), (3.16).
Proof. The Dirichlet problem defines a linear system of N equations in N

unknowns. From the comparison principle it follow that there is at most
one solution. From the linear algebra it is known that uniqueness implies
existence. ✷
Proposition 3.2.1 (A priori estimate). Assume u is a solution of the

discrete Dirichlet problem (3.15), (3.16), where c(x) ≤ 0. Then
max |u| ≤ max |Φ| + c max |f |,

Ωh ∂Ωh Ωh
where the constant c is independent of u and h, and 0 < h < h0 , h0 suffi-

ciently small, see the following proof for an explicit h0 .
Proof. The proof is the same as the proof of the apriori estimate of Theo-
rem 3.4. Concerning notations in the following formula see the proof of this
theorem. For a sufficiently large α we have
sinh(αh/2) 2

Lh eαd − eαx = −α2 eαx
αh/2
sinh(αh)
−b1 (x)αeαx + c(x) eαd − eαx
αh !
2
αx 2 sinh(αh/2) sinh(αh)
≤ −e α + b1 α .
αh/2 αh
Suppose that αh ≤ 1, then

Lh eαd − eαx ≤ −eαx α2 − Kα cosh(1)

≤ −1,
if we take an α which satisfies

q
α ≥ (K/2) cosh(1) + ((K/2) cosh(1))2 + 1.
We recall that we assume that Ω is bounded and is contained in the strip

0 < x1 < d, where d is the diameter of Ω. ✷
The following result says that the solution of the discrete Dirichlet problem
is an approximation of the solution of the original problem, provided that
this solution is sufficiently smooth,
Corollary. Suppose that u ∈ C 3 (Ω) is a solution of the continuous Dirichlet

problem (3.12), (3.13), where c(x) ≤ 0 and Φ is defined on a boundary
strip and is in C 1 in the closed strip. Let uh be a solution of the discrete
problem (3.15), (3.16). Then
max |u(x) − uh (x)| ≤ ch,

Ωh
where the constant c is independent on h < h0 , h0 sufficiently small.
Proof. Here we make the additional assumption that ∂Ωh ⊂ ∂Ω, see Fig-
ure 3.5 for an example. The proof of the general case is left as an exercise.
The assumption on u implies that
|Lh u − Lu| ≤ ch. (3.17)
Let U be defined on Ωh . Then we have from the above a priori estimate

that
max |U | ≤ max |U | + c max |Lh U |.
Ωh ∂Ωh Ωh
Set U = u − uh , then
max |u − uh | ≤ c max |Lh uh − Lh u|

Ωh Ωh
since u − uh = 0 on ∂Ωh . Finally, we have on Ωh that
Lh uh − Lh u = Lh uh − Lu + Lu − Lh u
= Lu − Lh u
since Lh uh = f and Lu = f on Ωh . Then the estimate of the corollary

follows from the estimate (3.17). ✷
3.3. EXERCISES 83
3.3 Exercises
1. Let Ω ⊂ Rn be a bounded domain and assume u ∈ C(Ω) and
sup (u + ǫeγx1 ) = sup (u + ǫeγx1 )
Ω ∂Ω
for each ǫ. Show that

sup u = sup u.
Ω ∂Ω
2. Let [aij ] be a real regular symmetric matrix in Rn . Find a regular

Matrix C in Rn such that
n
X n
X
aij uxi xj = U yi y i ,
i,j=1 i=1
where U (y) := u(C −1 y).
Hint: Let Z1 , Z2 , . . . , Zn be an orthonormal system of eigenvectors to

[aij ] to the eigenvalues λ1 , λ2 , . . . , λn , resp. Set B = (Z1 , Z2 , . . . , Zn ),
then  √ 
1/ λ1 0
√ . . . 0
 0 1/ λ2 0 ...   T
C=  ............................  B .
√
0 ... 0 1/ λn
3. Let γ be the interior angle of a sector with its corner at the origin in
R2 . Calculate the interior angle ω of the sector transformed by the
above mapping.
4. Let u ∈ C 1 (Ω) ∩ C 2 (Ω) be a solution of
∇u
div p = f inΩ
1 + |∇u|2
u = Φ on ∂Ω,
where Ω ∈ R2 and f, Φ are given. Suppose the origin is a corner of
Ω with interior angle γ, 0 < γ < π. Show that ω, see the previous
exercise, is the opening angle of the surface S defined by z = u(x1 , x2 ),
over the origin, see Figure 3.6.
Hint:
2
∇u X
div p = aij (x)uxi xj ,
1 + |∇u|2 i,j=1
ω S
γ
Ω
Figure 3.6: Transformed opening angle
where
u xi u xj

1
aij = δij − .
(1 + |∇u|2 )3/2 1 + |∇u|2
5. Let Ωα be the sector in R2 defined by r > 0 and 0 < θ < α, where

0 < α < 2π. Set Ωα,ρ = Ωα ∩ Bρ (0). Suppose that u ∈ C 2 (Ωα,ρ \ {0})
and supΩα,ρ |u| < ∞ is a solution of △u = 0 in Ωα,ρ , u = 0 on (∂Ωα ∩
Bρ (0)) \ {0}.
Show that there is a constant c such that
|u| ≤ c|x|π/α
in Ωα,ρ .
Hint: Choose the comparison function W = Arπ/α sin(πθ/α) and show
that −AW ≤ u ≤ AW on Ωα ∩ ∂Bρ (0) provided the positive constant
A is sufficiently large.
6. Let m, λj , cj be real numbers satisfying cj ≤ m, λj > 0 and 4j=1 λj =

P
1. Show that m ≤ 4j=1 λj cj implies cj = m.

P
7. Prove the corollary to Proposition 3.3.

Hint: The inclusion ∂Ωh ⊂ ∂Ω is not assumed. The result follows since
for given x ∈ ∂Ωh there exists an x1 ∈ ∂Ω such that |x − x1 | < h.
Bibliography
[1] R. A. Adams, Sobolev Spaces. Academic Press, New York, 1975.
[2] W. Blaschke, Vorlesungen über Differentialgeometrie I. Springer-

Verlag, Berlin 1921, Grundlehren, Bd 1.
[3] J. T. Chen, R. Finn and E. Miersemann, Capillary surfaces in wedge

domains. Pacific J. Math. 236 (2008), 283–305.
[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 Math-

ematics , Vol. 19, AMS, Providence, 1991.
[6] L. C. Evans and R. F. Gariepy, Measure Theory and Fine Properties of

Functions, Studies in Advanced Mathematics, CRC Press, Boca Raton,
1992.
[7] G. M. Fichtenholz, Differential-und Integralrechnung III. VEB

Deutscher Verlag der Wissenschaften, Berlin 1973. Translation from
Russian.
[8] R. Finn, Equilibrium Capillary Surfaces. Grundlehren, Vol. 284,

Springer-Verlag, New York, 1986.
[9] P. R. Garabedian, Partial Differential Equations. Chelsia Publishing

Company, New York, 1986.
[10] D. Gilbarg and N. S. Trudinger, Elliptic Partial Differential Equations

of Second Order. Grundlehren, Vol. 224, Springer-Verlag, Berlin, 1983.
85
86 BIBLIOGRAPHY
[11] E. Hopf, A remark on linear elliptic equations of second order. Proc

Amer. Math. Soc. 3 (1952), 791–793.
[12] E. Isaacson and H. B. Keller, Analysis of Numerical Methods. Wi-

ley&Sons, New York. German translation: Analyse numerischer Ver-
fahren., Edition Leipzig, 1972.
[13] F. John, Partial Differential Equations. Springer-Verlag, New York,

1982.
[14] S. G. Michlin, Lehrgang der Mathematischen Physik. Akademie-Verlag,

Berlin, 1975. Translation from Russian.
[15] E. Miersemann, Asymptotic expansions of solutions of the Dirichlet

problem for quasilinear elliptic equations of second order near a con-
ical point. Math. Nachr. 135 (1988), 239–274.
[16] E. Miersemann, Asymptotic expansion at a corner for the capillary

problem: the singular case. Pacific J. Math. 157 (1993), 95-107.
[17] E. Miersemann, Partial Differential Equations. Lecture Notes, Leipzig

University, 2009.
[18] E. Miersemann, Kapillarflächen. Ber. Verh. Sächs. Akad. Wiss. Leipzig,

Math.-Natur. Kl. 130 (2008), Heft 4, S. Hirzel, Leipzig, 2008.
[19] O. Perron, Eine neue Behandlung der ersten Randwertaufgabe für

△u = 0. Math. Z. 18, (1923), 42–54.
[20] I. G. Petrowski, Vorlesungen über Partielle Differentialgleichungen.

Teubner, Leipzig, 1955. Translation from Russian. Englisch translation:
Lectures on Partial Differential Equations. Wiley-Interscience, 1954.
[21] W. I. Smirnow, Lehrgang der Höheren Mathematik., Teil V. VEB Verlag

der Wiss., Berlin, 1975. Translation from Russian. English translation:
Course of Higher Mathematics, Vol. 2., Elsevier, 1964.
[22] W. A. Strauss, Partial Differential equations. An Introduction. Second

edition, Wiley-Interscience, 2008. German translation: Partielle Differ-
entialgleichungen. Vieweg, 1995.
[23] K. Yosida, Functional Analysis. Grundlehren, Vol. 123, Springer-Verlag,

Berlin, 1965.

Linear Elliptic Equations of Second Order: Lecture Notes

Uploaded by

Copyright:

Available Formats

Linear Elliptic Equations of Second Order: Lecture Notes

Uploaded by

Document Information

Original Description:

Original Title

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

Linear Elliptic Equations of Second Order: Lecture Notes

Uploaded by

Copyright:

Available Formats

Linear Elliptic Equations of Second Order

Version October, 2015

These lecture notes present an introduction to linear second order elliptic

The notation potential has its origin in Newton’s attraction rule

Here s(r) denotes the singularity function, see also [17],

We recall that ωn = |∂B1 (0)|. Since w ∈ C 2 (Ω) and −△w = f in Ω if f is

in Ω and v = h − w on ∂Ω. Consequently, it is sufficient to consider the

The Dirichlet problem (first boundary value problem) is to find a solution

where Φ is given and continuous on ∂Ω.

For the Dirichlet problems we make an ansatz with a dipole potential

if n ≥ 3. In the case that n = 2 we have to replace |z −y|2−n by − ln(|z −y|).

if n ≥ 3. In the case that n = 2 we have to replace |z −y|2−n by − ln(|z −y|).

Both potentials solve the Laplace equation in Rn \ ∂Ω.

In the rest of this chapter we assume that n ≥ 3.

Definition. We say that ∂Ω ∈ C 1,λ , 0 < λ ≤ 1, if:

f ∈ C 1,λ (D2ρ ), f (0) = 0, ∇f (0) = 0.

Figure 1.1: Definition of ∂Ω ∈ C 1,λ

that φi = 1 in Bρ (xi ) and 0 ≤ φi (x) ≤ 1, see an exercise. Set

which implies that

since at least one of the factors is zero. ✷

Assume ∂Ω ∈ C 1,λ , then we define the area integral by

Here we suppose that ∂Ω ∩ Z2ρ (xi ) has the parametric representation y =

1.2 Dipole potential

Figure 1.2: Double layer potential

charge of power ǫ−1 at y + = y + (ǫ/2)ν(y) and a charge of power −ǫ−1 at

Since z − = z + − ǫν(y), we have

Proof. Consider the case n ≥ 3. Let x be the center of a local coordinate

is continuous in a neighborhood of 0 ∈ Rn−1 . Here is D2ρ = D2ρ (0), z =

Figure 1.3: Local coordinates

Because of f ∈ C 1,λ (D2ρ ), f (0) = 0 and ∇f (0) = 0 we have

where A(ξ, ζ) is bounded on D2ρ × D2ρ and continuous if ξ 6= ζ. Since the

Let x0 ∈ ∂Ω and x ∈ Rn . Set

W (x) = W1 (x) + σ(x0 )W0 (x),

Lemma 1.2.2. Suppose that ∂Ω ∈ C 1,λ . Then

Proof. (i) If x ∈ Rn \ Ω is fixed, then there is a domain Ω0 ⊃⊃ Ω where

(ii) Let x ∈ Ω be fixed, then there is a ball Bρ (x) ⊂ Ω. Then

(iii) Let x ∈ ∂Ω and set for a sufficiently small ρ > 0,

Sρ = Ω ∩ ∂Bρ (x), Cρ = ∂Ω \ Bρ (x),

see Figure 1.4. Then

Since, see an exercise,

Figure 1.4: Figure to the proof of Lemma 1.2.2

Figure 1.5: Estimate of |Sρ |

Lemma 1.2.3. Let ∂Ω ∈ C 1,λ . Then

is uniformly bounded for x ∈ Rn .

Set Sd = ∂Ω ∩ Bd (x0 ). Then for y ∈ ∂Ω \ Sd we have

Figure 1.6: Local coordinates, case (iii)

where hi ∈ C(Dρ (0)), |hi (ξ)| ≤ c|ξ|1+λ .

|ξ|2 + |δ + h2 (ξ))2 = |ξ|2 + δ 2 + 2δh2 + h22

Id ≤ 2n/2 ωn−1 max{cλ−1 ρλ , π/2},

Proof. Set Sρ = ∂Ω ∩ Bρ (x0 ) and W1 (x) = I1 + I2 , where

Set, see Lemma 1.2.3,

and choose for given ǫ > 0 a ρ = ρ(ǫ) such that

provided that y ∈ ∂Ω \ Sρ . Since I2 is continuous in Bρ/2 (x0 ), there is a

if |x − x0 | < δ(ǫ). Summarizing, we have

|W1 (x) − W1 (x0 )| < ǫ

if |x − x0 | < min{ρ(ǫ)/2, δ(ǫ)}. ✷

Let x0 ∈ ∂Ω and denote by Wi (x0 ) the limit of W (x) from interior to x0

Proposition 1.2.1. Suppose that σ ∈ C(∂Ω) and x0 ∈ ∂Ω. The limits