Partial differential equations 2

Based on lectures by Claus Gerhardt

 Contents

1 Distributions and Sobolev spaces 2

1.1 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Sobolev-Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The difference quotient . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Sobolev embedding- and compactness theorems . . . . . . . . 19
1.5 L2 regularity for weak solutions . . . . . . . . . . . . . . . . . 30
1.6 Eigenvalueproblems for the Laplacian . . . . . . . . . . . . . . 36
1.7 The Harnack inequality . . . . . . . . . . . . . . . . . . . . . 43

2 Hölder continuity of weak solutions 53

2.1 Solution of the homogeneous equation . . . . . . . . . . . . . 53
2.2 Local Hoelder continuity . . . . . . . . . . . . . . . . . . . . . 55
2.3 Hoelder estimates near the boundary . . . . . . . . . . . . . . 63
2.4 Application to nonlinear equations . . . . . . . . . . . . . . . 68

3 Quasilinear operators and Leray-Schauder theory 75

3.1 Fixed point theorems, Leray-Schauder theorem and applications 75
3.2 Gradient bounds . . . . . . . . . . . . . . . . . . . . . . . . . 80

1
 Chapter 1

Distributions and Sobolev

spaces

1.1 Distributions
1.1.1 Definition. Let Ω ⊂ Rn be open, K ⊂ Ω compact. We set

DK (Ω) := {φ ∈ Cc∞ (Ω) : supp φ ⊂ K}.

On DK (Ω) we define the following norms:

∀m ∈ N : pm (φ) = |φ|m,K .

1.1.2 Remark. Those norms define a topology on DK (Ω), using the base

Um, := {φ : pm (φ) < },  > 0, m ∈ N,

such that DK (Ω) becomes a topological vector space, i.e., all the other neigh-
borhood bases are formed by translation. This topology is then generated
by the metric
X |φ − η|m
d(φ, η) := 2−m .
1 + |φ − η|m
m∈N

1.1.3 Proposition. T ∈ DK (Ω)∗ is continuous, if and only if

∃m ∈ N ∃c > 0 ∀φ ∈ DK (Ω) : |hT, φi| ≤ cpm (φ).

Proof. Exercise.
◦
1.1.4 Remark. Let Ki % Ω be an exhaustion, such that Ki ⊂K i+1 . Then
[
Cc∞ (Ω) = DKi (Ω) =: D(Ω).
i∈N

2
Let the topology T of D(Ω) be defined by the requirement
∀i ∈ N : T|DK (Ω) ⊂ TDK (Ω) .
i i

The topology T does not depend on the exhaustion.

Proof. Exercise.
1.1.5 Definition. (i) A linear form T on D(Ω) is called distribution, if it
is continuous. For the set of all continuous linear forms on D(Ω) we write
D0 (Ω).
(ii) D0 (Ω) obtains the ∗-weak topology, i.e.
∗
Ti * T ⇔ ∀φ ∈ D(Ω) : hTi , φi → hT, φi.
1.1.6 Remark. From the previous constructions we deduce
T ∈ D0 (Ω) ⇔ ∀K b Ω ∃m ∈ N ∃c > 0 ∀φ ∈ DK (Ω) : |hT, φi| ≤ cpm (φ).
If m can be chosen independently of K, the minimal such m is called order
of T , ord(T ).
1.1.7 Definition. A distribution of order 0 is called measure.
1.1.8 Remark. Let f ∈ L1loc (Ω), then
ˆ
hf, φi = fφ
Ω
defines a measure.
Proof. Exercise.
1.1.9 Definition. Let T ∈ D0 (Ω), α ∈ Nn . We define the α-th weak deriva-
tive or distributional derivative of T , Dα T by
hDα T, φi := (−1)|α| hT, Dα φi.
1.1.10 Remark. We have Dα T ∈ D0 (Ω) and ord(Dα T ) ≤ ord(T ) + |α|, if
both sides are defined.
1.1.11 Example. Let (
1, t>0
θ(t) :=
−1, t < 0
Then, as one easily verifies, θ0 = 2δ0 .
1.1.12 Remark. According to the fundamental lemma of the calculus of
variations,
Ψ : L1loc (Ω) ,→ D0 (Ω)
is an embedding.
The derivative Dα u of a function u ∈ L1loc (Ω) is always to be understood as
distributional derivative.
1.1.13 Remark. For Ψ(Lploc (Ω)) we simply write Lploc (Ω) and consider this
to be a subspace of D0 (Ω).

3
1.2 Sobolev-Spaces
1.2.1 Definition. Let n ≥ 1 and Ω ⊂ Rn be open, m ∈ N, 1 ≤ p ≤ ∞. By

H m,p (Ω) := {u ∈ Lp (Ω) : Dα u ∈ Lp (Ω) ∀|α| ≤ m}

 1
p
X
α
kukm,p =  kD ukpp  , 1 ≤ p < ∞,
|α|≤m
X
kukm,∞ = kDα uk∞ ,
|α|≤m

we denote the space of Sobolev functions of class (m, p). On H m,2 (Ω) we
define the scalar product
X ˆ
hu, vi := Dα uDα v.
|α|≤m Ω

1.2.2 Remark. H m,p (Ω) is complete for 1 ≤ p ≤ ∞.

Proof. Exercise.

1.2.3 Lemma. (i) Let u ∈ H m,p (Rn ), 1 ≤ p < ∞ and (η ) be a Dirac
sequence, then we have for
ˆ
u (x) = η (x − y)u(y)dy
Rn

(a) ∀|α| ≤ m : Dα u = (Dα u)

(b) u → u in H m,p (Rn )

(ii) Let Ω0 b Ω ⊂ Rn be open and u ∈ H m,p (Ω), 1 ≤ p < ∞. Extend u to Rn

by 0. Then
u → u in H m,p (Ω0 ),  < dist(Ω0 , ∂Ω).

Proof. Exercise.

1.2.4 Lemma. (Product rule)

0 1 1
Let f ∈ H 1,p (Ω) and g ∈ H 1,p (Ω), 1 ≤ p ≤ ∞ and p + p0 = 1. Then

f · g ∈ H 1,1 (Ω)

and
D(f g) = Df · g + f · Dg.

4
Proof. By symmetry we may assume p < ∞. Extend f, g to Rn by 0 and let
f be the mollified sequence as in 1.2.3. Let ζ ∈ Cc∞ (Ω). Then there holds
ˆ ˆ
(ζf )∂i g = − (ζ∂i f g + f ∂i ζg).
Ω Ω

Taking the limit  → 0 via Hoelder’s theorem we obtain

ˆ ˆ
∞
∀ζ ∈ Cc (Ω) : ζ(f ∂i g + ∂i f g) = − f g∂i ζ.
Ω Ω

Again by Hoelder’s inequality we obtain

D(f g) ∈ L1 (Ω).

1.2.5 Lemma. (Chain rule)

Let Ω b Rn , g ∈ C m (R) and |g|m ≤ c. Then for u ∈ H m,p (Ω) we have
g ◦ u ∈ H m,p (Ω) and
D(g ◦ u) = g 0 (u)Du.
Proof. Let m = 1 and 1 ≤ p < ∞. Let φ ∈ Cc∞ (Ω) and Ω0 b Ω, such that
φ ∈ Cc∞ (Ω0 ). Let u ∈ C ∞ (Ω̄0 ) such that

ku − u km,p,Ω0 → 0

and
(u , Du ) → (u, Du) a.e.
ˆ ˆ  ˆ 
0
⇒ (g ◦ u)Di φ = lim (g ◦ u )Di φ = lim − g (u )Di u φ (1.1)
Ω0 →0 Ω0 →0 Ω0
There holds g 0 (u ) → g 0 (u) a.e. and |g 0 | ≤ L.

⇒ |φg 0 (u )Du| ≤ L|Du||φ|.

Dominated convergence implies

ˆ ˆ
|g 0 (u )Di u φ − g 0 (u)Di uφ| ≤ |g 0 (u )(Di u − Di u)φ|
0 0
Ω ˆΩ
+ |g 0 (u ) − g 0 (u)||Di u||φ| → 0.
Ω0

(1.1) implies the chain rule. Furthermore we have

1
kg ◦ uk1,p,Ω0 ≤ ckuk1,p,Ω + c|Ω| p

⇒ g ◦ u ∈ H 1,p (Ω).
From this estimate we deduce, using p → ∞, the claim for p = ∞. For m > 1
use induction and the product rule.

5
1.2.6 Theorem. Let x̃ ∈ Diff m (Ω, Ω̃) such that x̃ and x̃−1 have a bounded
C m −norm and 1 ≤ p ≤ ∞.
Then the map
Φ : H m,p (Ω) → H m,p (Ω̃)
u 7→ ũ = u ◦ x̃−1
is a topological isomorphism.
Proof. We show this for m = 1, the rest follows by induction.
Let Ω0 b Ω, u ∈ H 1,p (Ω), u → u in H 1,p (Ω0 ).
ũ = u ◦ x̃−1
∂xk
⇒ D̃i ũ = Dk u .
∂ x̃i
Let the sequence also satisfy
ũ → ũ a.e.
and
∂xk
D̃i ũ → Dk u a.e.
∂ x̃i
By the transformation theorem and the boundedness of the Jacobians we
have
ũ ∈ H 1,p (Ω̃0 )
and
∀Ω0 b Ω : kũk1,p,Ω̃0 ≤ ckuk1,p,Ω0
⇒ kũk1,p,Ω̃ ≤ ckuk1,p,Ω .
By symmetry this also holds for the inverse. For p = ∞ the claim holds by
taking the limit.

1.2.7 Lemma. Let u ∈ H 1,p (Ω), then

u+ = max(u, 0), u− = min(u, 0) and |u|
are in H 1,p (Ω) and a.e. there holds
(
+ Du, u>0
Du =
0, u≤0
(
Du, u<0
Du− =
0, u≥0
and 
Du,
 u>0
D|u| = 0, u=0.

−Du, u<0


6
Proof. Let  > 0. (√
t2 + 2 − , t > 0
g (t) :=
0, t ≤ 0.
Then g ∈ C 1 and |g0 | ≤ 1.

g → max(·, 0) locally uniformly.

The chain rule implies

u := g ◦ u ∈ H 1,p (Ω)

and (
√uDu , u>0
Du = g0 (u)Du = u2 +2
0, u ≤ 0.
Let η ∈ Cc∞ (Ω).
ˆ ˆ
u Di η = − Di u η
Ω ˆΩ
uD u
=− √ i η
{u>0} u2 + 2
ˆ ˆ
uD u
=− √ i χ{u>0} η → − χ{u>0} Di uη.
Ω u2 + 2 Ω

Since the left hand side converges to

ˆ
u+ Di η,
Ω

we obtain the claim. Using

u− = −(−u)+

and
|u| = u+ − u−
the other cases also follow.

1.2.8 Corollary. Let u ∈ H 1,p (Ω), c ∈ R, E := {u = c}.

⇒ Du|E = 0 a.e.

Proof. Wlog c = 0. There holds u = u+ +u− . Apply the previous lemma.

7
1.2.9 Theorem. Let Ω b Rn , u ∈ H 1,p (Ω) and let g ∈ C 0,1 (R) such that
Lip(g) ≤ L and suppose g 0 has only at most countably many points of dis-
continuity. Let M be the set of those points. Then

v := g ◦ u ∈ H 1,p (Ω)

and we have (
g 0 (u)Du, u(x) ∈
/M
Dv =
0, u(x) ∈ M.

Proof. Let g be a mollification of g

⇒ g → g locally uniformly

and
g0 → g 0 locally uniformly in M c ,
as well as
|g0 | ≤ L.
Then
v := g ◦ u ∈ H 1,p (Ω)
and
Dv = g0 (u)Du.
Let M = {tk : k ∈ H ⊂ N} and
[
Ek := {u = tk }, E := Ek .
k∈H

⇒ Du|E = 0 a.e.
There holds g 0 (u)Du ∈ Lp (Ω) and for a.e. x ∈ Ω we have
(
g 0 (u(x))Du(x), x ∈
/E
lim g0 (u(x))Du(x) =
→0 0, x ∈ E.

1.2.10 Remark. This theorem also holds for arbitrary g ∈ C 0,1 (R), |g 0 | ≤ L,
c.f. Ziemer: Weakly differentiable functions.

1.2.11 Theorem. Let Ω b Rn be open, ∂Ω ∈ C 0,1 . Then there holds

ˆ p ˆ ˆ
1 2
∀u ∈ C (Ω̄) : |u| ≤ 1 + L |Du| + c |u|,
∂Ω Ω Ω

where L is an upper bound for the Lipschitz constants of the boundary rep-
resentations.

8
Proof. (i) Let x0 ∈ ∂Ω and φ be a local graph representation around 0 ∈
Rn−1 ,
Γ = {(x̂, φ(x̂)) : |x̂| < ρ}.
Furthermore let 0 < a, such that

U = {(x̂, xn ) : φ(x̂) < xn < a, x̂ ∈ B̂ρ (0)} ⊂ Ω.

For a function u having support in this chart we then have

ˆ ˆ p p ˆ
|u| = |u(x̂, φ(x̂))| 1 + |Dφ|2 ≤ 1 + L2 |u|.
Γ B̂ρ (0) B̂ρ (0)

Also suppose, that u(·, a) = 0. Then

ˆ φ(x̂)
u(x̂, φ(x̂)) = Dn u(x̂, t)dt
a
ˆ a ˆ a
⇒ |u(x̂, φ(x̂))| ≤ |Dn u| ≤ |Du|.
φ(x̂) φ(x̂)

ˆ p ˆ
⇒ |u| ≤ 1 + L 2 |u(x̂, φ(x̂))|
Γ B̂ρ (0)
ˆ ˆ a p
≤ |Du| 1 + L2
B̂ρ (0) φ(x̂)
p ˆ
= 1+L 2 |Du|.
U

(ii) Now consider an open covering (Bρi ), 1 ≤ i ≤ N, of ∂Ω, such that ∂Ω ∩

Bρi can be represented as a graph locally and also such that the conditions
of (i) are satisfied.
Let (ηi ) be a subordinate finite partition of unity for ∂Ω. Then
N
X
u= uηi on ∂Ω.
i=1

ˆ N ˆ
X N p
X ˆ
⇒ |u| ≤ |uηi | ≤ 1+L2 |D(uηi )|
∂Ω i=1 ∂Ω i=1 Ω

p ˆ XN p ˆ N
X
≤ 1 + L2 |Du| ηi + 1 + L2 |u| |Dηi |
Ω i=1 Ω i=1
p ˆ ˆ
≤ 1+L2 |Du| + c |u|.
Ω Ω

9
1.2.12 Remark. ´ ´ ´
(i) ∂Ω ∈ C 1 ⇒ ∀u ∈ C 1 (Ω̄) : ´∂Ω |u| ≤ (1
´ + ) Ω |Du|
´ + c Ω |u|
2 1
(ii) ∂Ω ∈ C ⇒ ∀u ∈ C (Ω̄) : ∂Ω |u| ≤ Ω |Du| + c Ω |u|.

Proof. Exercise

1.2.13 Definition. We say Ω satisfies the H m,p - extension property, if there

exists Ω ⊂ Ω0 b Rn and a continuous linear map

F : H m,p (Ω) → H0m,p (Ω0 ),

such that
∀u ∈ H m,p (Ω) : F u|Ω = u.
F is then called extension operator.

1.2.14 Definition. Let E ⊂ Rn be measurable. Then the Sobolev spaces

H m,p (E) and H0m,p (E) respectively are defined as the closure of

{u ∈ C m (E) : kukm,p,E < ∞}

and Ccm (E) respectively with respect to the norm k · km,p .

1.2.15 Theorem. Let Ω b Rn be open and ∂Ω ∈ C m , then there holds for

1≤p<∞
H m,p (Ω̄) = H m,p (Ω).

Proof. First choose a local boundary neighborhood U , such that 1.2.6 implies

H m,p (U ) = H m,p (B1+ (0)).

Let u ∈ Hcm,p (B1+ (0) ∪ {xn = 0}). Define

uh (x̂, xn ) := u(x̂, xn + h), h > 0.

Then uh is defined in B1+ (0) − hen . For small  = (h) we then find

uh, = uh ∗ η ∈ C ∞ (B1+¯(0)).

Later we will show, that

kuh − ukm,p → 0, h → 0.

Thus we find
uhk ,k → u in H m,p (B1+ (0)).

⇒ u ∈ H m,p (B1+ (0)).

Using a partition of unity we obtain the claim. The other inclusions follow
immediately from the definitions.

10
1.2.16 Lemma. (Lions-Magenes)
Let c1 , ..., cm+1 be solutions of the system
m+1
X
(−1)j k j ck = 1, 0 ≤ j ≤ m.
k=1

Then
m+1
X
n
ũ(x̂, x ) = ck u(x̂, −kxn ), xn < 0
k=1

defines an extension for u ∈ C m (R¯n+ ) ∩ H m,p (Rn+ ) into all of Rn , such that

ũ ∈ C m (Rn )

and
kũkm,p,Rn ≤ ckukm,p,Rn+ , c = c(m, n, p), 1 ≤ p ≤ ∞.

Proof. Exercise

1.2.17 Corollary. Let Ω b Rn be open and ∂Ω ∈ C m . Then Ω satisfies the

H m,p - extension property for all 1 ≤ p < ∞.

Proof. Clear by the previous theorem and lemma.

1.2.18 Remark. (i) Ω b Rn ⇒ H0m,p (Ω) ,→ Hcm,p (Rn ).

(ii) ∂Ω ∈ C 0,1 ⇒ Ω satisfies the H m,p extension property (Calderon-Zygmund,
without proof).
(iii) For 1 ≤ p < ∞, ∂Ω ∈ C 0,1 ⇒ H m,p (Ω) = H m,p (Ω̄).

Proof. (i) is clear and (iii) follows from (ii) immediately.

1.2.19 Theorem. Let Ω b Rn be open, ∂Ω ∈ C 0,1 . Then there exists a

continuous trace operator

t : H 1,p (Ω) → Lp (∂Ω), 1 ≤ p < ∞,

such that
t|H 1,p (Ω)∩C 0 (Ω̄) = ·|∂Ω .

Proof. Since we have H 1,p (Ω) = H 1,p (Ω̄), it suffices to prove the claim for
u ∈ C ∞ (Ω̄).
(i) For u ∈ C 1 (Ω̄) define t(u) = u|∂Ω . We have
ˆ p ˆ ˆ
|u| ≤ 1 + L 2 |Du| + c |u|,
∂Ω Ω Ω

11
which also holds for Lipschitz functions by approximation. We apply this
estimate to |u|p yielding
ˆ p ˆ ˆ
p p−1
|u| ≤ p 1 + L 2 |Du||u| + c |u|p
∂Ω Ω Ω
ˆ  1 ˆ  p−1 ˆ
p p
≤ c0 |Du|p |u|p +c |u|p
Ω Ω Ω

⇒ kt(u)kp,∂Ω ≤ ckuk1,p,Ω .
(ii) Let u ∈ H 1,p (Ω) and

u = u ∗ η ∈ Cc∞ (Rn )

⇒ u → u in H 1,p (Ω̄).
⇒ kt(u )kp,∂Ω ≤ cku k1,p,Ω .
Thus we can define
t(u) := lim t(u ).
→0

(iii) Let u ∈ H 1,p (Ω) ∩ C 0 (Ω̄). We may suppose u ∈ H 1,p (Rn ) ∩ C 0 (Rn ).

t(u ) → t(u) in Lp (∂Ω)

and
u → u in C 0 (Ω̄)
imply the claim.

1.2.20 Proposition. u ∈ H01,p (Ω) ⇒ t(u) = 0.

Proof. Follows immediately from the preceding proof.

1.2.21 Proposition. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . Let m ≥ 1,

1 ≤ p < ∞. Then for u ∈ H m,p (Ω) all the Dβ u, |β| ≤ m − 1, are defined on
∂Ω in the sense of traces.

Proof. All those functions are in H 1,p (Ω).

1.2.22 Proposition. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . For u, v ∈

H 1,p (Ω) there holds

t(max(u, v)) = max(t(u), t(v))

t(min(u, v)) = min(t(u), t(v)).

Proof. By approximation.

12
1.2.23 Lemma. Let Ω b Rn , ∂Ω ∈ C 0,1 , u ∈ H 1,p (Ω), 1 ≤ p < ∞. Then
we have for large k
ˆ ˆ ˆ
p p p
(i) k p |u|p ≤ cp k p−1 1 + L2 |u|p + c 1 + L2 (|Du|p + |u|p ).
Ω1 ∂Ω Ω√
k 1+L2
ˆ p ˆ ˆ k

(ii) |u| ≤ k 1 + L2 |u| + c (|Du| + |u|)

∂Ω Ω1 Ω1
ˆ p ˆ k k
ˆ
2 2
(iii) lim sup k |u| ≤ 1 + L |u| ≤ (1 + L ) lim inf k |u|
k→∞ Ω1 ∂Ω k→∞ Ω1
k
ˆ k

(iv) t(u) = 0 ⇒ lim sup k p |u|p = 0,

k→∞ Ω1
k

where (
1, if p = 1
cp =
c(p, ∂Ω), if p > 1,
Ωk = {x ∈ Ω : d(x, ∂Ω) < k} and d = dist(·, ∂Ω).

Proof. (i) Let u ∈ C 1 (Ω̄), wlog supp(u) ∩ Ω̄ ⊂ B̂R (0) × (0, a) =: G. Let
1
k < min(a, R), then
 
n n 1
Ω 1 ∩ G = (x̂, x ) ∈ Ω : |x̂| < R ∧ d(x̂, x ) < .
k k
p
∀(x̂, xn ) ∈ Ω 1 ∩ G ∃ŷ ∈ B̂2R (0) : d(x̂, xn ) = |x̂ − ŷ|2 + |φ(ŷ) − xn |2 ,
k

where ∂Ω ∩ B̂2R (0) × (0, a) = graph φ. Thus for all (x̂, xn ) ∈ Ω 1 ∩ G we have
k

⇒ |xn − φ(x̂)| ≤ |xn − φ(ŷ)| + |φ(ŷ) − φ(x̂)|

≤ |xn − φ(ŷ)| + L|x̂ − ŷ|
p p
≤ 1 + L2 |xn − φ(ŷ)|2 + |x̂ − ŷ|2
p
≤ k −1 1 + L2

 
n 1p 2 n
⇒ Ω1 ∩ G ⊂ (x̂, x ) : |x̂| < R, φ(x̂) < x < φ(x̂) + 1+L .
k k
ˆ xn
n
|u(x̂, x ) − u(x̂, φ(x̂))| ≤ |Dn u(x̂, t)|dt.
φ(x̂)

|u(x̂, xn )| ≤ |u(x̂, xn ) − u(x̂, φ(x̂))| + |u(x̂, φ(x̂))|

|u(x̂, xn )|p ≤ 2p (|u(x̂, xn ) − u(x̂, φ(x̂))|p + |u(x̂, φ(x̂))|p ).

13
Set (
1, if p = 1
cp = −1 p
p 2 , if p > 1.
Then we find
√ √
ˆ ˆ φ(x̂)+ 1+L2 ˆ ˆ φ(x̂)+ 1+L 2
k p p k
n p −p
|u(x̂, x )| ≤ 1 + L cp k2 |Dn u|p
B̂R φ(x̂) B̂R φ(x̂)
p ˆ
−1
+ cp k 1+L 2 |u(x̂, φ(x̂))|p
B̂R
√
ˆ ˆ φ(x̂)+ 1+L2
p p k
≤ cp k −p 1 + L2 |Du|p
B̂R φ(x̂)
p ˆ
−1
+ cp k 1+L 2 |u|p .
∂Ω

Furthermore we have
√
n n 1 + L2
{(x̂, x ) ∈ Ω : |x̂| < R, φ(x̂) < x < φ(x̂) + } ⊂ Ω √1+L2 .
k k

√
ˆ ˆ ˆ φ(x̂)+ 1+L2
k
p
⇒ |u| ≤ |u(x̂, xn )|p
Ω 1 ∩G B̂R φ(x̂)
k
ˆ
p p
−p
≤ cp k 1+L2 |Du|p (1.2)
Ω√
1+L2

p ˆ k

+ cp k −1 1 + L2 |u|p .
∂Ω

(ii) From

|u(x̂, φ(x̂))| ≤ |u(x̂, xn ) − u(x̂, φ(x̂))| + |u(x̂, xn )|

ˆ xn
≤ |u(x̂, xn )| + |Dn u(x̂, t)|dt
φ(x̂)

we deduce
ˆ ˆ ˆ φ(x̂)+ k1 ˆ ˆ φ(x̂)+ k1
−1 n −1
k |u(x̂, φ(x̂))| ≤ |u(x̂, x )| + k |Du|.
B̂R B̂R φ(x̂) B̂R φ(x̂)

ˆ ˆ p
−1 −1
k |u| ≤ k |u(x̂, φ(x̂))| 1 + L2
∂Ω B̂R
p ˆ p ˆ (1.3)
≤ 1 + L2 |u| + k −1 1 + L2 |Du|.
Ω1 Ω1
k k

14
This also holds for all u ∈ H 1,p (Ω) sucht that supp(u) ∩ Ω̄ ⊂ G.
Let u ∈ H 1,p (Ω) and consider a covering of Ω¯1 by ui , 1 ≤ i ≤ N , together
k0
with a subordinate partition of unity (ηi ), such that (1.2) and (1.3) are
applicable to uηi . Thus
ˆ ˆ ˆ
p p p
|u|p ≤ ck −p 1 + L2 (|Du|p + |u|p ) + cp k −1 1 + L2 N p |u|p
Ω1 Ω1 ∂Ω
k k

and ˆ p ˆ ˆ
|u| ≤ 1 + L2 k |u| + c (|Du| + |u|).
∂Ω Ω1 Ω1
k k

(iii) and (iv) follow from (i) and (ii) easily.

1.2.24 Lemma. Let Ω b Rn be open and ∂Ω ∈ C 0,1 , 1 ≤ p < ∞. Let

u ∈ H 1,p (Ω), t(u) = 0. Then there holds

u ∈ H01,p (Ω).

Proof. d = dist(·, ∂Ω) ∈ C 0,1 (Rn ) and |Dd| = 1 a.e. Set

ηk := min(1, kd), k ≥ 1.

Let Ωk be the corresponding boundary strip. Then we find

ηk = 1 in Ω\Ω 1 .
k

(i) Claim: u ∈ H 1,p (Ω) ⇒ uηk ∈ H01,p (Ω).

Proof: Let u ∈ C 0,1 (Ω̄)

⇒ v := uηk ∈ C 0,1 (Ω̄) ∧ uηk|∂Ω = 0.

Let  > 0 and using a decomposition into v + and v − we may as well suppose
v ≥ 0.
v := max(v − , 0) ∈ Cc0,1 (Ω) ⊂ H01,p (Ω),
which follows from approximation. We have
(
Dv, if v > 
Dv =
0, if v ≤ .
ˆ ˆ
|Dv − Dv |p = |Dv|p → 0, since |Ω| < ∞.
Ω {v≤}
ˆ ˆ ˆ
p p
|v − v | =  1+ |v|p → 0.
Ω {v>} {v≤}

15
Let u ∈ H 1,p (Ω), t(u) = 0. Then for a mollification u we have
u → u in H 1,p (Rn )
⇒ u ηk → uηk in H 1,p (Ω).
⇒ uηk ∈ H01,p (Ω).
(ii) Furthermore we have
ˆ ˆ ˆ
|Du − D(uηk )| ≤ k |u| + |Du| → 0,
Ω Ω1 Ω1
k k

by the preceding lemma.

p > 1: ˆ ˆ ˆ
|Du − D(uηk )|p ≤ 2p |Du|p + 2p k p |u|p .
Ω Ω1 Ω1
k k

Analogously ˆ ˆ
p
|u − uηk | ≤ |u|p .
Ω Ω1
k

Thus u ∈ H01,p (Ω).

1.2.25 Proposition. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . Let u ∈ H 1,p (Ω),

t(u) ≤ k a.e. on ∂Ω. Then
max(u − k, 0) ∈ H01,p (Ω).
Proof. t(max(u−k, 0)) = max(t(u)−k, 0) = 0 and use the preceding lemma.

1.2.26 Corollary. Let Ω b Rn be open and ∂Ω ∈ C 1 , u ∈ H 1,1 (Ω). Then

ˆ ˆ
k |u| → |u|.
Ω1 ∂Ω
k

Proof. For C 1 boundary it is possible to obtain L ≤  for all  > 0.

1.2.27 Lemma. For h ∈ Rn , v ∈ Lp (Rn ), 1 ≤ p < ∞ define
vh (x) = v(x + h).
(i) This defines an isometry of Lp (Rn ), kvkp = kvh kp ,
(ii) limh→0 kv − vh kp = 0 and
(iii) For Ω ⊂ Rn and Lp (Ω) → Lp (Rn ) extending by zero we have
kvh kp,Ω ≤ kvkp,Ω
and
kvh − vkp,Ω → 0.
Proof. Exercise.

16
1.3 The difference quotient
In this chapter we consider for a given function u the so-called difference
quotient
u(x + hen ) − u(x)
∆h u(x) = , 0 6= h ∈ R.
h
Abusing notation, let
h = hen .
1.3.1 Lemma. Let Ω ⊂ Rn be open. For Ω0 b Ω and h < dist(Ω0 , ∂Ω) we
have that
∆h : Lp (Ω) → Lp (Ω0 )
is continuous and
k∆h ukp,Ω0 ≤ 2|h|−1 kukp,Ω .
Furthermore there holds

h∆h u, viL2 = −hu, ∆−h viL2 ,

if one of the functions has compact support in Ω and h is small.

Proof. W.l.o.g. let supp(v) ⊂ Ω and Ω0 = int(supp(v)). Then we have
ˆ
u(x + h) − u(x)
h∆h u, vi = v(x)dx
Ωˆ0 h
ˆ
1 1
= u(x + h)v(x)dx − u(x)v(x)dx
h Ω0 h Ω0
ˆ
v(y) − v(y − h)
= − u(y) dy
h
ˆ Ω
v(y − h) − v(y)
= − u(y) dy
Ω −h
= −hu, ∆−h vi.

1.3.2 Lemma. (i) Let Ω b Rn be open, u ∈ H 1,p (Ω), 1 ≤ p < ∞, Ω0 b Ω.

Then
∀|h| < h0 << 1 : k∆h ukp,Ω0 ≤ kDn ukp,Ω (1.4)
and
lim kDn u − ∆h ukp,Ω0 = 0. (1.5)
h→0

(ii) For u ∈ H 1,p (Rn ) there hold

k∆h ukp,Rn ≤ kDn ukp,Rn (1.6)

and
k∆h ukp,Rn → kDn ukp,Rn . (1.7)

17
Proof. Let Ω0 b Ω00 b Ω and h < dist(∂Ω, Ω00 ).
(i) Since we can approximate u by u ∈ C 1 (Ω) ∩ H 1,p (Ω) and since (∆h u) =
∆h u we have
∆h u → ∆h u in H 1,p (Ω0 ),
as  → 0. Thus let u ∈ C 1 (Ω) ∩ H 1,p (Ω). Let x ∈ Ω0 b Ω, h > 0.
ˆ
1 xn +h
∆h u(x) = Dn u(x̂, t)dt,
h xn
thus
ˆ xn +h p
p −p
|∆h u(x)| ≤ h Dn u(x̂, t)dt
xn
ˆ xn +h
−p p−1
≤h h |Dn u(x̂, t)|p dt
xn
ˆ xn +h
= h−1 |Dn u(x̂, t)|p dt.
xn

Thus we have
ˆ ˆ hˆ
|∆h u(x)|p dx ≤ h−1 |Dn u(x̂, xn + t)|p dxdt ≤ kDn ukpp,Ω .
Ω0 0 Ω0

For −h this holds, since ∆−h u(x) = ∆h u(x − h). Let  > 0. Choose v ∈
C 1 (Ω) ∩ H 1,p (Ω) such that

kv − uk1,p,Ω0 < .
3
Then

kDn u − ∆h ukp,Ω0 ≤ kDn u − Dn vkp,Ω0 + kDn v − ∆h vkp,Ω0 + k∆h (u − v)kp,Ω0 .

The first and last term are less than 3 . The middle term´s integrand con-
verges to 0 uniformly.
(ii) The proof is exactly the same, but instead of the uniform convergence in
the last argument use the decomposition
ˆ ˆ ˆ
p p
|Dn v − ∆h v| ≤ |Dn v − ∆h v| + |Dn v − ∆h v|p
Rn BR |x|>R

and that the functions are integrable.

1.3.3 Lemma. Let Ω b Rn be open, u ∈ H m,p (Ω), 1 < p < ∞, m ∈ N,

Ω0 b Ω and let

∀|α| ≤ m : k∆h Dα ukp,Ω0 ≤ c ∀|h| ≤ h0 .

18
Then
Dn u ∈ H m,p (Ω0 )
and
kDn Dα ukp,Ω0 ≤ c.
Proof. 1 < p < ∞ ⇒ Lp (Ω0 ) is reflexive. Thus there exists a sequence hk
such that
∆hk Dα u * vα ∈ Lp (Ω0 )
and
kvα kp,Ω0 ≤ lim inf kDhk Dα ukp,Ω0 ≤ c.
k→∞

Let η ∈ Cc∞ (Ω0 ). Then

hvα , ηi = lim h∆hk Dα u, ηi = (−1)|α|+1 hu, Dn Dα ηi.

k→∞

Thus, if |α| = 0 we have Dn u = vα .

If |α| ≥ 1, we have Dn u ∈ H m,p (Ω0 ).

1.4 Sobolev embedding- and compactness theorems

1.4.1 Theorem. Let Ω b Rn be open with H 1,p -extension property,
1 ≤ p < n. Then there holds
∗
H 1,p (Ω) ,→ Lp (Ω),
1 1
where p∗ = p − n1 .
Proof. We show

∃c = c(n, p) ∀u ∈ H 1,p (Ω) : kukp∗ ≤ ckuk1,p .

It suffices to show this for u ∈ Cc∞ (Rn ). Let first be p = 1 and x = (x̂i , xi )
for all i. ˆ xi
|u(x)| ≤ |Di u(x̂i , t)|dt
−∞
n ˆ ∞  1
n Y n−1
⇒ |u(x)| n−1 ≤ |Di u(x̂i , t)|dt
i=1 −∞

ˆ ∞ ˆ ∞  1
n−1
n
1
⇒ |u| n−1 dx ≤ |D1 u(x̂1 , t)|dt
−∞ −∞
ˆ n ˆ ∞
∞ Y  1
n−1
· |Di u(x̂i , t)|dt dx1
−∞ i=2 −∞

19
The generalized Hoelder inequality implies
ˆ ∞ ˆ ∞  1
n−1
n
1
⇒ |u| n−1 dx ≤ |D1 u(x̂1 , t)|dt
−∞ −∞
n ˆ ∞ ˆ ∞  1
Y n−1
i i 1
· |Di u(x̂i , x )|dx dx .
i=2 −∞ −∞

For n = 2 this already implies

ˆ ∞ˆ ∞ ˆ ∞ ˆ ∞  ˆ ∞ ˆ ∞ 
n
|u| n−1 ≤ |D1 u| |D2 u| .
−∞ −∞ −∞ −∞ −∞ −∞

For n > 2 we repeat this argument to obtain

ˆ ∞ ˆ ∞ ˆ ∞ ˆ ∞  1
n−1
n
1 2 2 2 1
|u| n−1 dx dx ≤ |D2 u(x̂2 , x )|dx dx
−∞ −∞ −∞ −∞
ˆ ∞ ˆ ∞  1
n−1
1 1 2
· |D1 u(x̂1 , x )|dx dx
−∞ −∞
n ˆ ∞ ˆ ∞ ˆ ∞  1
Y n−1
i i
· |Di u(x̂i , x )|dx
i=3 −∞ −∞ −∞

Successive integration implies

ˆ n ˆ  1
n−1
ˆ  n
n−1
n Y
|u| n−1 ≤ |Di u| ≤ |Du|
Rn i=1 Rn Rn

⇒ ∀u ∈ Cc∞ (Rn ) : kuk n−1

n ≤ kDuk1 .
Let now 1 < p < n : Define

p(n − 1)
t := > 1, u ∈ Cc∞ (Rn )
n−p

⇒ v := |u|t ∈ Cc1 (Rn )

ˆ ˆ  n
n−1
n
⇒ |v| n−1 ≤ |Dv| .
Rn Rn
n(p−1)
|Dv| ≤ t|u| n−p |Du|
ˆ n(p−1)
ˆ
np p−1
⇒ kvk n−1
n ≤t |u| n−p |Du| ≤ tkDukp ( |u| n−p ) p

Rn Rn
⇒ kukp∗ ≤ tkDukp .

20
1.4.2 Corollary. For u ∈ H01,p (Ω) there even holds

kukp∗ ≤ ckDukp ,

which also means, that kDukp,Ω is a norm on H01,p (Ω).

Proof. This follows from the extension property, i.e.

H 1,p (Ω) ,→ H01,p (Ω0 ) ,→ Hc1,p (Rn )

and the previous proof.

1.4.3 Theorem. Suppose Ω has the H m,p - extension property. Then

H m,p (Ω) ,→ Lq (Ω),

1 1 m
q = p − n, if mp < n.

Proof. Exercise.

1.4.4 Proposition. Let Ω have the H m,p - extension property and |Ω| < ∞.
Let mp = n ≥ 2. Then

∀1 ≤ q < ∞ : H m,p (Ω) ,→ Lq (Ω).

Proof. (i) p > 1: Let p −  > 1. Then

H m,p (Ω) ,→ H m,p− (Ω)

and
m(p − ) < n.
Thus
H m,p− (Ω) ,→ Lq (Ω),
where q → ∞.
(ii) p = 1: Then m ≥ 2 and for u ∈ H m,1 (Ω) we have Dm−1 u ∈ H 1,1 (Ω) ,→
n
L n−1 (Ω). Thus
n
H n,1 (Ω) ,→ H n−1, n−1 (Ω).
Now (i) is applicable.

1.4.5 Remark. 1.4.4 does not hold for q = ∞.

Proof. Choose Ω = B 1 (0) ⊂ Rn , n ≥ 2 and

2

u(x) = log(− log |x|) − log log 2.

There holds
1 1 x
Du =
log |x| |x| |x|

21
and
ˆ ˆ 1
n n−1
2 1 1 n−1
|Du| = |S | − r
Ω 0 logn r rn
ˆ 1
2 1
= |S n−1
| r−1
| logn r|
ˆ ∞
0
1
=c dt < ∞.
log 2 tn

1.4.6 Theorem.
H m,p (Rn ) = H0m,p (Rn ),
if 1 ≤ p < ∞.

Proof. We only prove the case m = 1, the rest follows from induction. Let
0 ≤ η ≤ 1, η ∈ Cc∞ (Rn ), such that
(
1, |x| ≤ 1
η(x) =
0, |x| ≥ 2

and
|Dη| ≤ c.
Set x
ηk (x) = η .
k
For u ∈ H 1,p (Rn ) define

uk = uηk ∈ H01,p (Rn ).

There clearly holds uk → u in Lp (Rn ).

Furthermore Duk = Duηk + k −1 uDη → Du in Lp (Rn ).

1.4.7 Theorem. Let Ω b Rn have the H 1,p - extension property. Let p > n,
then for α = 1 − np we have

H 1,p (Ω) ,→ C 0,α (Ω̄)

and
∀u ∈ H01,p (Ω) : [u]α,Ω ≤ ckDukp .

Proof. Without loss of generality let u ∈ H01,p (Ω0 ), Ω b Ω0 , and we will

show
∀u ∈ H01,p (Ω0 ) : |u|0,α,Ω0 ≤ ckDukp .

22
Let x1 , x2 ∈ Ω0 , 0 < ρ = |x1 − x2 |, x ∈ Bρ ( x1 +x
2 ) ≡ Bρ (0). Then we have
2

1
for u ∈ Cc (Ω0 )
ˆ 1
d
u(x) − u(xi ) = u(xi + t(x − xi ))dt
0 dt
ˆ 1
≡ Dk u(xt )(xk − xki )dt
0
ˆ 1
≤ 2ρ |Du(xt )|.
0

Thus
ˆ 1ˆ
1−n
u − u(xi ) ≤ 2cρ |Du(xi + t(x − xi ))|
Bρ 0 Bρ
ˆ 1 ˆ
1−n −n
≤ 2cρ t |Du(z)|
0 B2ρt (xi )
ˆ 1
n p−1 n p−1
≤ 2cρ 1−n
t−n kDukp,Ω0 ρ p t p

0
ˆ 1
1− n −n
≤ cρ p kDukp,Ω0 t p
0
1− n
≤ c(n, p)kDukp,Ω0 ρ p .

Finally

|u(x1 ) − u(x2 )| ≤ u(x1 ) − u + u − u(x2 )

Bρ Bρ

≤ ckDukp |x1 − x2 |α .

Choosing x2 ∈ ∂Ω0 we find u(x2 ) = 0 and thus

|u|0,Ω0 ≤ ckDukp (diamΩ)α .

1.4.8 Theorem. Let Ω b Rn have the H m,p - extension property. Then

H m,p (Ω) ,→ C j,α (Ω̄), m ∈ N, 1 ≤ p < ∞,

if

m = k + j and
n
(i) (k − 1)p < n < kp, α = k −
p
(ii) (k − 1)p = n, ∀0 < α < 1.

23
Proof. Exercise.

1.4.9 Theorem. (Interpolation theorem)

Let 1 ≤ p1 < p < p2 < ∞, p1 = pα1 + 1−α
p2 , 0 < α < 1 and Ω be a measure
space. Then

∀u ∈ Lp1 (Ω) ∩ Lp2 (Ω) : kukp ≤ kukαp1 kukp1−α

2
.

Proof. There holds

1
p= (αp1 p2 + (1 − α)p1 p2 ).
αp2 + (1 − α)p1
Thus
ˆ ˆ αp2 (1−α)p1
p1 αp p2 αp
|u|p = |u| 2 +(1−α)p1 |u| 2 +(1−α)p1
Ω Ω
ˆ  αp2 ˆ  (1−α)p1
αp2 +(1−α)p1 αp2 +(1−α)p1
p1 p2
≤ |u| |u| .
Ω Ω

1.4.10 Theorem. (Kolmogorov)

Let Ω b Rn . A subset M ⊂ Lp (Ω), 1 ≤ p < ∞, is precompact if and only if

(i) M is bounded and

(ii) M is equicontinuous in the mean,

i.e.
∀ > 0 ∃δ > 0 ∀u ∈ M : 0 ≤ h < δ ⇒ ku − uh kp,Ω < .

Proof. Let M be precompact. Then M is clearly bounded. Let  > 0. Then

there exist (ui )1≤i≤N such that
N
[
M⊂ B (ui ).
i=1

Let u ∈ M , then u ∈ B (ui0 ).

⇒ ku(· + h) − ukp,Ω ≤ ku(· + h) − ui0 (· + h)k

+ kui0 (· + h) − ui0 k + kui0 − uk < 3,

if we choose h small enough. Note that a finite collection of functions is

equicontinuous.
Now let (i) and (ii) hold. Let  > 0 and for δ > 0 let ηδ be a Dirac sequence.
Let
uδ = u ∗ ηδ .

24
 ˆ p
p
|uδ (x) − u(x)| = ηδ (y)(u(x − y) − u(x)) dy
Bδ (0)
ˆ
≤ ηδ (y)|u(x − y) − u(x)|p dy
Bδ (0)
ˆ ˆ ˆ
p
⇒ |uδ − u| ≤ ηδ (y) |u(x − y) − u(x)|p dxdy
Rn Bδ (0) Rn

(ii) ⇒ kuδ − ukp ≤ sup ku(x − y) − u(x)kp < ,

|y|<δ

if δ is small.
We now claim that Mδ := {uδ : u ∈ M } ⊂ C 0 (Ω + δ) =: E is precompact in
E. We have
ˆ
1− 1 1
|uδ (x)| ≤ ηδ p (y)ηδp (y)|u(x − y)|dy
Bδ (0)
ˆ !1
p

≤ ηδ (y)|u(x − y)|p
Bδ (0)
1
≤ sup |ηδ | p kukp ≤ c
Bδ

Thus Mδ is bounded.
Furthermore
ˆ
1− p1 1
|uδ (x + h) − uδ (x)| ≤ ηδ ηδp |u(x + h − y) − u(x − y)|dy
Bδ (0)
1
≤ sup |ηδ | p ku(y + h) − u(y)kp .
Bδ (0)

Thus Mδ is equicontinuous and by Arzela-Ascoli there exists an -net

(uiδ )1≤i≤N in E. We now claim, that this net is also an -net in Lp (Ω). Let
u ∈ M and 1 ≤ i ≤ N. Then
ˆ ˆ ˆ
i p p p p
|u − uδ | ≤ 2 |u − uδ | + 2 |uδ − uiδ |p ≤ cp .
Rn Rn Rn

1.4.11 Proposition. (Kondrašov)

1 1
Let Ω b Rn have the H 1,p - extension property, 1 ≤ p < ∞. Let p∗ = p − n1 ,
then for q < p∗
H 1,p (Ω) ,→ Lq (Ω)
is compact.

25
Proof. Let u ∈ H 1,p (Ω) be bounded. Suppose

∀ : u ∈ H 1,p (Ω0 )

and
ku k1,p,Ω0 ≤ c.
⇒ ∀ > 0 ∃v ∈ Cc∞ (Ω0 ) : kv − u k < .
Thus it suffices to show, that the v are precompact in Lq (Ω0 ). By the
interpolation theorem this will follow from the case q = 1. We use the
Kolmogorov characterization. The boundedness is clear.
ˆ 1
d
v (x + h) − v (x) = v (x + th)dt
0 dt
ˆ 1
= Di v (x + th)hi dt
0

and thus
ˆ ˆ 1ˆ
|v (x + h) − v (x)| ≤ |h| |Dv | ≤ |h|kDv k1 .
Rn 0 Rn

1 1 m
1.4.12 Corollary. Let Ω have the H m,p - extension property, q > p − n,
q ≥ 1. Then
H m,p (Ω) ,→ Lq (Ω)
is compact. In cases mp = n this holds for all 1 ≤ q < ∞.
Proof. The case m = 1 has been proven. There holds

u, Du ∈ H m−1,p (Ω) ,→ Lr (Ω),

where
1 1 m−1
= − .
r p n
Thus u ∈ H 1,r (Ω) ,→ Lq (Ω), being compact, if
1 1 1 1 1 m
> ∗ = − = − .
q r q n p n
The second claim follows by interpolation.

1.4.13 Lemma. (Interpolation of Hoelder spaces)

Let Ω b Rn be open and 0 < β < α ≤ 1. Then there holds
β β
[u]β,Ω ≤ [u]αα · (osc(u))1− α
β β
β 1− α
≤ [u]αα · 21− α |u|0 .

26
Proof.
α !β
α
|u(x) − u(y)| |u(x) − u(y)| β
=
|x − y|β |x − y|α
 β
|u(x) − u(y)| α
−1
α
= β
|u(x) − u(y)| β
|x − y|
β β
≤ [u]α,Ω
α
(osc(u))1− α .

1.4.14 Corollary. Let Ω b Rn be open and ∂Ω ∈ C 0,1 , 0 < β < α. Then

the embedding
C k,α (Ω̄) ,→ C k,β (Ω̄)
is compact.
Proof. Let u ∈ C k,α (Ω̄) be bounded. By Arzela-Ascoli there exists a subse-
quence
u → u ∈ C k,α (Ω̄) in C k (Ω̄).
Set
v := Dγ u → Dγ u = v
for some multiindex γ. Inserting this into the interpolation theorem yields
the result.

1.4.15 Theorem. Let Ω b Rn be open and ∂Ω ∈ C 0,1 , mp > n. Then

H m,p (Ω) ,→ C j,β (Ω̄), 0 ≤ β < α,
is compact, where j, α are as in the Sobolev embedding theorem.
1.4.16 Lemma. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . Then
C 0,1 (Ω̄) = H 1,∞ (Ω).
Proof. Let u ∈ C 0,1 (Ω̄). Then a mollification u converges in C 0,1 (Ω̄0 ) to u
for all Ω0 b Ω. Thus u ∈ H 1,∞ (Ω). Let u ∈ H 1,∞ (Ω). Since
|u(x) − u(y)| ≤ kDuk∞,Ω |x − y|,
we obtain the result locally. For x, y ∈ Bδ (x0 ) ∩ Ω, x0 ∈ ∂Ω, we can use a
coordinate transformation to convert the problem into the convex set B+1 (0).

1.4.17 Proposition. Let Ω b Rn be open and ∂Ω ∈ C 0,1

⇒ H m,p (Ω) ,→ H m−1,p (Ω), 1 ≤ p < ∞, m ≥ 1,
is compact.

27
Proof. Follows immediately from the other embedding theorems.

1.4.18 Proposition. Let Ω b Rn be open and ∂Ω ∈ C 0,1 , m ≥ 1 and

1 ≤ p < ∞. Then
X
∀ > 0 ∃c ∈ R ∀u ∈ H m,p (Ω) : kukm−1,p,Ω ≤  kDα ukp,Ω + c kukp,Ω .
|α|=m

Proof. Use the compactness lemma for Banach spaces and absorb the lower
order norm in the left hand side.

1.4.19 Corollary. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . Then the norm

X
kuk = kDα ukp,Ω + kukp,Ω , 1 ≤ p < ∞,
|α|=m

is an equivalent norm on H m,p (Ω).

1.4.20 Lemma. Let Ω b Rn . Then

kuk = kDm ukp,Ω

is an equivalent norm on H0m,p (Ω).

Proof. ∀|γ| ≤ m − 1 : Dγ u ∈ H01,p (Ω).

1.4.21 Theorem. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . Then the embedding

H 1,p (Ω) ,→ Lq (∂Ω)

(n−1)p
is compact for 1 < p < n and 1 ≤ q < n−p and it is continuous for
(n−1)p
q= n−p .

Proof. Let kuk k1,p,Ω ≤ c. Then a subsequence converges in L1 (Ω),

uk → u ∈ L1 (Ω).

Since, by reflexivity, we have u ∈ H 1,p (Ω) we may assume u ≡ 0.

Let  > 0.
ˆ ˆ ˆ
|uk | ≤ |Duk | + c |uk |
∂Ω Ω Ω
ˆ 1 ˆ
p p−1
p
≤c |Duk | |Ω | p + c |uk |
Ω Ω

Thus ˆ
p−1
lim sup |uk | ≤ c|Ω | p → 0,  → 0.
k→∞ ∂Ω

28
 ⇒ H 1,p (Ω) ,→ L1 (∂Ω)
(n−1)p
is compact. Let q = n−p and set v := |u|q ∈ H 1,1 (Ω)
n(p−1)
⇒ |Dv| ≤ |Du||u| n−p

ˆ ˆ  1 ˆ
p
 p−1
p
np
p
⇒ |Dv| ≤ |Du| |u| n−p

Ω Ω Ω
p(q−1)
≤ ckuk1,p,Ω .

⇒ H 1,p (Ω) ,→ Lq (∂Ω).

1.4.22 Theorem. (Poincare-inequality)

Let Ω b Rn be connected with H 1,p - extension property, 1 ≤ p < n. Then
for all measurable subsets E ⊂ Ω, |E| > 0, there exists a constant cE > 0,
such that
ˆ 1 ˆ 1
p p
1,p p p
∀u ∈ H (Ω) : |u − uE | ≤ cE |Du| ,
Ω Ω
1
´
where uE = |E| E u.
Proof. Set  ˆ 
1,p
V := u∈H (Ω) : u=0 .
E
Suppose the inequality did not hold, then there existed a sequence uk ∈ V
such that
kuk k1,p,Ω = 1
and
kuk kp,Ω > kkDuk kp,Ω .
By compactness we have a subsequence converging to u ∈ Lp (Ω). Thus
kDuk = 0
and so u ≡ const, which is a contradiction.
1.4.23 Theorem. For Ω ⊂ Rn open, the spaces H m,p (Ω) are reflexive for
1 < p < ∞.
Proof. Exercise.
1.4.24 Theorem. Let Ω ⊂ Rn be open and 1 ≤ p < ∞. Then
 
X 0

H0m,p (Ω)∗ ≡ H −m,p (Ω) = Dγ fγ : fγ ∈ Lp (Ω) ⊂ D(Ω).
 
|γ|≤m

Proof. Exercise.

29
1.5 L2 regularity for weak solutions
1.5.1 Theorem. (Interior estimates)
Let Ω b Rn and let ai ∈ C 1 (Ω × R × Rn ) satisfy

∂ai
∀(x, u, p) : (x, u, p) ≤ cA (1 + |u| + |p|) (1.8)
∂x
∂ai ∂ai
+ ≤c (1.9)
∂u ∂pj

and
∂ai
aij = ⇒ ∃λ > 0 ∀ξ ∈ Rn : λ|ξ|2 ≤ aij ξi ξj . (1.10)
∂pj
1,2
Let u ∈ Hloc (Ω) be a weak solution of the equation

Au = −(ai (x, u, Du))i = f ∈ L2 (Ω),

i.e. we have equality in H −1,2 (Ω). Then we have

2,2
u ∈ Hloc (Ω)

and for all Ω0 b Ω00 b Ω

kuk2,2,Ω0 ≤ c(kf k2,Ω , kuk1,2,Ω00 , cA , λ).

Proof. We use the method of difference quotients. Let h = hek for a fixed
1 ≤ k ≤ n. Let h0 be small enough to ensure Ω0 + h b Ω00 for all |h| ≤ h0 .
Let η ∈ Cc∞ (Ω00 ), such that
η|Ω0 = 1.
Multiply the equation by

−∆−h (∆h uη 2 ) ∈ H01,2 (Ω)

to obtain
ˆ ˆ
i 2
∆h (a (x, u, Du))(∆h uη )i = − f ∆−h (∆h uη 2 ).
Ω Ω

We have

∆h ai (x, u, Du) = h−1 (ai (x + h, u(x + h), Du(x + h)) − ai (x, u(x), Du(x)))
ˆ 1
−1 d i
=h a (x + th, tu(x + h) + (1 − t)u(x), ...)dt
0 dt
= h−1 (aij (u(x + h) − u(x))j + bi (u(x + h) − u(x)) + ci h),

30
where ˆ ˆ ˆ
1 1 n 1
ij ∂ai ∂ai X ∂ai
a = , bi = , ci = .
0 ∂pj 0 ∂u 0 ∂xk
k=1
By the assumptions we have

|ci | ≤ cA (1 + |u(x)| + |Du(x)| + |h||∆h u| + |h||∆h Du|),

|aij | + |bi | ≤ c
as well as the uniform ellipticity of aij . There holds
ˆ ˆ
(aij (∆h u)j + bi ∆h u + ci )(∆h uη 2 )i = − f ∆−h (∆h uη 2 )
Ω ˆΩ ˆ
 1
≤ f2 + |D(∆h uη 2 )|2 ,
2 Ω00 2 Ω

ˆ ˆ
ij 2
a (∆h u)j (∆h uη )i = aij (∆h u)j (∆h u)i η 2
Ω ˆ
Ω

+2 aij (∆h u)j ηi ∆h uη.

Ω

We have
ˆ ˆ

aij (∆h u)j ηηi ∆h u ≤ aij (∆h u)j (∆h u)i η 2
2 Ω
Ω
ˆ (1.11)
1
+ aij ηi ηj |∆h u|2 .
2 Ω
But ˆ ˆ
ij 2
a ηi ηj |∆h u| ≤ c(Dη) |Dk u|2 ,
Ω Ω00
ˆ ˆ
i 2
| b ∆h u(∆h uη )i | ≤ |bi ||∆h u|(|D∆h u|η 2 + 2|∆h u||Dη|η) (1.12)
Ω Ω
and
ˆ ˆ
i 2
| c (∆h uη )i | ≤ (1 + |u| + |Du| + |h||∆h u| + |D∆h u||h|)
Ω Ω (1.13)
2
· (|D∆h u|η + 2|∆h u||Dη|η).

For small  we obtain, also absorbing the |D∆h u| in (1.12) and (1.13),
ˆ ˆ
λ 2 1
|D∆h u| ≤ aij (∆h u)i (∆h u)j η 2
2 Ω0 2 Ω
ˆ
≤c (|f |2 + |Du|2 + |u|2 + 1) ∀|h| < h0 .
Ω00

31
 ˆ ˆ
2 2
⇒ |DDk u| ≤ c (f 2 + |Du|2 + |u|2 + 1)
Ω0 λ Ω00
2,2
⇒ u ∈ Hloc (Ω).

1.5.2 Remark. Now we want to prove boundary estimates. Since a diver-

gence writes in coordinates
1 ∂ √ i
−aii = − √ ( ga )
g ∂xi

we even may suppose that the differential operator is given in terms of co-
√
variant derivatives, after possibly multiplying the right hand side by g and
√
the vector filed by g −1 . Thus we are given a function on both sides and are
free to consider the equation on B1+ (0) without loss of generality.

1.5.3 Theorem. (Local boundary estimates)

Let 0 < ρ1 < ρ2 < ρ, x0 ∈ ∂Ω and Bρ (x0 ) ∩ ∂Ω = Γ ∈ C 2 . Let u ∈ H 1,2 (Ω)
be a solution of

−(ai (x, u, Du))i = f, u|∂Ω = φ ∈ H 2,2 (Ω),

where ai satisfies (1.8), (1.9) and (1.10).Then

u ∈ H 2,2 (Ω ∩ Bρ1 (x0 ))

and
kuk2,2,Ωρ1 ≤ c(kuk1,2,Ωρ2 , kf k2,Ω , kφk2,2,Ωρ2 , cA , ρ1 , ρ2 , |Γ|2 ),
where Ωρi = Ω ∩ Bρi (x0 ).

Proof. Without loss of generality the equation holds in Ω = B1+ (0) with
x0 = 0. Choose
0 ≤ η ∈ Cc∞ (Bρ2 ), η|Bρ1 ≡ 1.
Define with abuse of notation

h = h · ek , 1 ≤ k ≤ n − 1.

Multiply the equation with

−∆−h (∆h (u − φ)η 2 ) ∈ H01,2 (Ω).

Then ˆ ˆ
i 2
∆h a Di (∆h (u − φ)η ) = − f ∆−h (∆h (u − φ)η 2 ).
Ω Ω

32
As in the proof of 1.5.1 we obtain
ˆ ˆ ˆ ˆ !
2 2 2 2 2 2
λ |D∆h u| η ≤ c f + |D∆h φ| + 1 + (|Du| + u )
Ω Ωρ2 Ωρ2 Ωρ2

ˆ X ˆ ˆ ˆ !
2 2 2 2 2
⇒ |Di Dj u| ≤ c f + |D∆h φ| + 1 + (|Du| + u ) .
Ωρ1 i+j<2n Ωρ2 Ωρ2 Ωρ2

−Di ai (x, u, Du) = f

∂ai ∂ai
⇒ −aij uij − − ui = f.
∂xi ∂u
Using ann ≥ λ, we obtain the claim.
m,2 1,2
1.5.4 Theorem. Let aij , bi , c ∈ C m,1 (Ω), f ∈ Hloc (Ω) and u ∈ Hloc (Ω) be
a weak solution of
− (aij uj )i + bi ui + cu = f, (1.14)
then
m+2,2
u ∈ Hloc (Ω)
and for all Ω0 b Ω00 b Ω we have

kukm+2,Ω0 ≤ c(kf km,2,Ω00 + kuk1,2,Ω00 ),

where c = c(|aij |m,1,Ω00 , |bi |m,1,Ω00 , |c|m,1,Ω00 , Ω0 , Ω00 ).

Proof. By induction. For m = 0 this is theorem 1.5.1 So let m > 0 and

1,2
suppose the claim holds for m − 1. For 1 ≤ k ≤ n choose v = uk ∈ Hloc (Ω).

∂aij
 
ij i
⇒ −(a vj )i + b vi + cv = fk + uj − bjk uj + ck u ≡ F ∈ Hloc
m−1,2
(Ω).
∂xk i

Let Ω0 b Ω̃ b Ω00 .

⇒ kvkm+1,2,Ω0 ≤ c(kF km−1,2,Ω̃ + kvk1,2,Ω00 )

kvk1,2,Ω̃ ≤ kuk2,2,Ω̃ ≤ c(kf k2,Ω00 + kuk1,2,Ω00 )

and

kF km−1,2,Ω̃ ≤ c(kf km,2,Ω00 + kukm+1,2,Ω̃ ) ≤ c(kf km,2,Ω00 + kuk1,2,Ω00 ).

33
1.5.5 Theorem. (Local boundary estimates of higher order)
Let 0 < ρ1 < ρ2 < ρ, , x0 ∈ ∂Ω, Bρ (x0 ) ∩ Ω = Γ ∈ C m+2 . Let u ∈ H 1,2 (Ω)
be a solution of
−(aij uj )i + bi ui + cu = f, u|∂Ω = φ ∈ H m+2,2 (Ω),
f ∈ H m,2 (Ω), aij , bi , c ∈ C m,1 (Ω ∩ Bρ (x0 )). Then
kukm+2,2,Ωρ1 ≤ c(kf km,2,Ωρ2 + kuk1,2,Ωρ2 + kφkm+2,2,Ωρ2 ),

where c = c(|aij |m,1,Ω00 , |bi |m,1,Ω00 , |c|m,1,Ω00 , Ω0 , Ω00 ).

Proof. By induction, where m = 0 has already been proven. Let m > 0 and
suppose without loss of generality Ω = B1+ (0), x0 = 0. Set
Γ = B1 (0) ∩ {xn = 0}.
Let 1 ≤ k ≤ n − 1 and
v = uk ∈ H 1,2 (Ωρ ), v|Γ = φk ∈ H m+1,2 (Ωρ ).
Then
∂aij
−(aij vj )i + bi vi + cv = fk + uj − (bj uj )k − ck u ≡ F ∈ H m−1,2 (Ωρ2 ).
∂xk
Let 0 < ρ1 < ρ̃ < ρ2
⇒ kvkm+1,2,Ωρ1 ≤ c(kF km−1,2,Ωρ̃ + kvk1,2,Ωρ̃ + kφkm+2,2,Ωρ̃ ).
For k = n we again use the differential equation to obtain
n−1
X
kunn km,2,Ωρ1 ≤ c(kukm+1,2,Ωρ̃ + kuk km+1,2,Ωρ̃ + kφkm+2,2,Ωρ̃ + kf km,2,Ωρ̃ ).
k=1

We now consider L2 -estimates for the Neumann boundary value problem.

1.5.6 Theorem. Let Ω b Rn be open, ∂Ω ∈ C 2 and let u ∈ H 1,2 (Ω) be a
weak solution of
−(ai (x, u, Du))i = f in Ω, −ai νi = φ on ∂Ω,
where f ∈ L2 (Ω), φ ∈ H 2,2 (Ω) or φ ∈ C 0,1 (∂Ω), ai ∈ C 1 (Ω̄ × R × Rn ) and
let (1.8), (1.9) as well as (1.10). Then we have u ∈ H 2,2 (Ω) and
kuk2,2,Ω ≤ c(kφk2,2,Ω + kf k2,Ω + kuk1,2,Ω + 1)
in case φ ∈ H 2,2 (Ω) and
kuk2,2,Ω ≤ c(kf k2,Ω + kuk1,2,Ω + 1),
where c now also depends on |φ|0,1,∂Ω .

34
Proof. We only prove the boundary estimates, since the interior estimates
are theorem 1.5.1. Let Ω = B1+ (0), Γ = {xn = 0} ∩ ∂Ω. Then the weak
formulation of the equation reads
ˆ ˆ ˆ
1,2 1,2 i
∀η ∈ H (Ω) ∩ Hc (B1 (0)) : a ηi + φη = f η.
Ω Γ Ω

Let 1 ≤ k ≤ n − 1, h = h · ek be small and

η̃ = −∆−h (∆h uη 2 ), η ∈ Cc1 (B1 (0)).

Then
ˆ ˆ ˆ
i 2 2
∆h a (∆h uη )i + ∆h φ∆h uη = − f ∆−h (∆h uη 2 ).
Ω ∂Ω Ω

(i) If φ ∈ C 0,1 (∂Ω), we have

ˆ ˆ
|∆h φ∆h uη | ≤ L |∆h uη 2 |
2
Γ
ˆΓ ˆ
≤ L |D(∆h uη 2 )| + c |∆h uη 2 |,
Ω Ω

which can be absorbed by  in the left hand side.

(ii) If φ ∈ H 2,2 (Ω), we have
ˆ ˆ ˆ
2
|∆h φ∆h uη | ≤ |D(∆h φ∆h uη )| + c |∆h φ∆h uη 2 |.
2
Γ Ω Ω

1.5.7 Theorem. Let Ω b Rn be open, ∂Ω ∈ C 2 and let aij , bi , c ∈ L∞ (Ω),

c ≥ c0 > 0, aij uniformly elliptic, f ∈ L2 (Ω) and φ ∈ H 1,2 (Ω). Then

−(aij uj )i + bi ui + cu = f in Ω
−aij uj νi = φ on ∂Ω

has a weak solution u ∈ H 1,2 (Ω).

If additionally ∂Ω ∈ C m+2 , aij ∈ C m+1 (Ω), bi , c ∈ C m (Ω), f ∈ H m,2 (Ω)
and φ ∈ C m,1 (∂Ω) or φ ∈ H m+2,2 (Ω), then we have

u ∈ H m+2,2 (Ω)

and
kukm+2,2,Ω ≤ c(kφkm+2,2,Ω + kf km,2,Ω + kuk1,2,Ω ),
if φ ∈ H m+2,2 (Ω). If φ ∈ C m,1 (∂Ω), then the constant also depends on
|φ|m,1,∂Ω .
Proof. Exercise.

35
1.6 Eigenvalueproblems for the Laplacian
In this section we want to solve the eigenvalue problems

−∆u = λu in Ω
(1.15)
u|∂Ω = 0,

−∆u = λu in Ω
∂u (1.16)
=0
∂ν
and
− ∆u = λu in M, (1.17)
where M is a compact Riemannian manifold.
We will reduce each of these problems to an abstract eigenvalue problem in
a suitable Hilbert space.
1.6.1 Assumptions of this section. In this section we use the following
assumptions:
(1) H is a real, separable Hilbert space.
(2) K is a symmetric, continuous and compact bilinear form on H, such that

∀u 6= 0 : K(u) = K(u, u) > 0.

(3) B is a symmetric, continuous bilinear form on H, which is coercive

relative K, i.e.

∃c0 , c > 0 ∀u ∈ H : B(u) = B(u, u) ≥ ckuk2 − c0 K(u).

We will solve the abstract eigenvalue problem

∃0 6= u ∈ H, λ ∈ R ∀v ∈ H : B(u, v) = λK(u, v).

1.6.2 Lemma. Let {0} =

6 V ⊂ H be a closed subspace. Then the variational
problem
B(v) → min, v ∈ W := V ∩ {K(v) = 1}
has a solution u, which is also a solution of
B(v)
→ min, 0 6= v ∈ V.
K(v)
Setting
B(v)
λ = inf ,
06=v∈V K(v)
then we have
∀v ∈ V : B(u, v) = λK(u, v).

36
Proof. By coercivity we see, that B is bounded below in W and that a
minimal sequence u is bounded above. Thus we suppose

u + u ∈ V.

⇒ K(u ) → K(u) = 1.
B is lower semicontinuous, because B + c0 K is an equivalent norm on H.
Thus the first two claims follow. The eigenvalue problem is the first variation
of
B(v)
v 7→ .
K(v)

1.6.3 Theorem. The eigenvalue problem

∀v ∈ H : B(ui , v) = λi K(ui , v)

has countably many eigenvalues of finite multiplicity. If we write

λ1 ≤ λ2 ≤ ...,

we find
lim λi = ∞.
i→∞

The eigenvectors (ui ) are complete in H. They fulfill the orthogonality rela-
tions
K(ui , uj ) = δij
and
B(ui , uj ) = λi K(ui , uj ),
as well as the expansions
X
B(u, v) = λi K(ui , u)K(ui , v)
i

and X
K(u, v) = K(ui , u)K(ui , v).
i

The pairs (λi , ui ) are defined by the variational problem

 
B(u)
λi = B(ui , ui ) = inf : 0 6= u ∈ H, K(u, uj ) = 0 ∀1 ≤ j ≤ i − 1 .
K(u)

37
Proof. 1. Solve the variational problem

B(u)
→ min, 0 6= u ∈ H.
K(u)

By the previous theorem there exists a solution u1 and there holds

∀v ∈ H : B(u1 , v) = λ1 K(u1 , v), K(u1 ) = 1,

such that λ1 is the infimum.

2. Let i > 1 and let there be solutions for 1 ≤ j ≤ i − 1. Set

Vi = hu1 , ..., ui−1 i

and let V ⊥ be the orthogonal complement of V relative K. Again, by the

previous theorem
 
⊥ B(u) ⊥
∃ui ∈ V : B(ui ) = λi = inf :u∈V
K(u)

and
∀v ∈ V ⊥ : B(ui , v) = λi K(ui , v).
For 1 ≤ j ≤ i − 1 we have

B(uj , ui ) = λj K(uj , ui ) = 0.

Thus
∀v ∈ H : B(ui , v) = λi K(ui , v),
since
H = Vi ⊕K Vi⊥ .
Let u ∈ H and set
m
X
um = K(u, ui )ui ∈ Vm+1 .
i=1
⊥
⇒ u = um + (u − um ) ∈ Vm+1 ⊕ Vm+1 .
The ui satisfy the orthogonality relation

B(ui , uj ) = λi K(ui , uj ) = λδij .

3. Suppose now the eigenvalues were bounded. We have

B(ui ) = λi

and
K(ui ) = 1,

38
and thus
c0 K(ui ) + B(ui ) = λi + c0 ,
so that
kui k ≤ c.
⇒ 2 = K(ui − ui+1 ) → 0
for a subsequence, which is a contradiction. By the same reasoning the
multiplicity must be finite.
4. We prove the completeness. Let u ∈ H.
m
X m
X
ũm = K(u, ui )ui ≡ ci ui .
i=1 i=1

Set
vm = u − ũm .
⊥
vm ∈ Vm+1
and thus
λm+1 K(vm ) ≤ B(vm ).
m
X
K(vm ) = K(u) − c2i
i=1
and
m
X
B(vm ) = B(u) − λi c2i
i=1
imply
B(vm ) ≤ c
and thus
K(vm ) → 0.
Furthermore there holds
∞
X
λi c2i < ∞.
i=1
Let m < n.
n
X
B(vn − vm ) = λi c2i < .
i=m+1
Thus the (vn ) form a Cauchy sequence in H and by K(vm ) → 0 we find
vm → 0.
Thus the (ui ) are complete and
∞
X
B(u) = λi c2i .
i=1

39
1.6.4 Theorem. (Minimax principle)
For a subspace V ⊂ H define
 
B(u)
d(V ) = inf : 0 6= u ∈ V ⊥ .
K(u)
Then λi is characterized by

λi = max{d(V ) : V ⊂ H, dimV ≤ i − 1}

where the maximum is attained at

hu1 , ..., ui−1 i,

where the ui are defined as in 1.6.3.

Proof. For i ≥ 2 let
Vi = hv1 , ..., vi−1 i.
For i = 1 the claim has already been proven. We show

d(Vi ) ≤ λi = d(hu1 , ..., ui−1 i).

Set
i
X
u= cj uj , cj ∈ R
j=1

and solve
K(u, vj ) = 0 1 ≤ j ≤ i − 1.
Let u be a solution with K(u) = ij=1 c2j = 1.
P

i
B(u) X
d(Vi ) ≤ = λj c2j ≤ λi .
K(u)
j=1

1.6.5 Example. Let Ω b Rn and consider (1.15). This eigenvalue problem

is realized in the above setting by

H = H01,2 (Ω),
ˆ
B(u, v) = Di uDi v
Ω
and ˆ
K(u, v) = uv.
Ω
Those bilinear forms obviously satisfy the assumptions of the abstract eigen-
value problem. Furthermore we have:

40
1.6.6 Theorem. The smallest eigenvalue, λ1 , has multiplicity 1 and a cor-
responding eigenfunction u1 has a strict sign.

Proof. Exercise.

1.6.7 Example. Let Ω b Rn be open and ∂Ω ∈ C 0,1 , H = H 1,2 (Ω). Con-

sider (1.16). This eigenvalue problem is realized by setting
ˆ
B(u, v) = Di uDi v
Ω

and ˆ
K(u, v) = uv.
Ω

1.6.8 Example. To solve (1.17) we define the bilinear forms as in the pre-
vious examples on the space H = H 1,2 (M ).

1.6.9 Definition. Let f : M → R be a function.

(a) f is called measurable on M , if f is measurable in coordinates.
(b) We say f ∈ Lp (M ), if f is measurable and
ˆ
|f |p < ∞.
M

(c) Let u ∈ Lp (M ) and (η i ) ∈ Cc∞ (M, Rn ). Define the weak derivative of

first order of u, (Di u), to be a tensor satisfying
ˆ ˆ
i
Di uη = − u div η.
M M

(d) Let
ˆ X
   
 m 
X p
H m,p (M ) = u ∈ Lp (M ) :  |Dα uDα u| 2  < ∞ .
 M k=0 |α|=k


1.6.10 Lemma. Let u ∈ C 2 (M ). Then −∆ is the Euler-Lagrange operator

of the functional ˆ
1
J(v) = |Dv|2 .
2 M
Proof. ˆ
∀η ∈ Cc∞ (M ) : 0 = δJ(u; η) = ui η i .
M

1.6.11 Theorem. Let Ω b M, then the Sobolev embedding theorems also

hold for H m,p (Ω) and H0m,p (Ω).

41
Proof. The case m = 1 is an exercise and the rest follows by induction.

1.6.12 Theorem. Let M be compact. Then there are countably many eigen-
values λi of −∆,
0 = λ0 < λ1 ≤ λ2 ≤ ... → ∞.
The eigenfunctions are complete in L2 (M ) as well as in H 1,2 (M ). The kernel
of −∆ is spanned by a nonzero constant function.
Proof. The claim follows from the above examples and by 11.8.16, Analysis
II.

1.6.13 Theorem. Let u be harmonic and homogeneous of degree k in a

neighborhood of Sn . Then u|Sn is an eigenfunction with eigenvalue λ = k(k +
n − 1) of −∆Sn .
Let u be an eigenfunction with eigenvalue λ = k(k + n − 1) on Sn of −∆Sn ,
then we have
u ∈ C ∞ (Sn ).
In Rn+1 define  
x
u(x) = u |x|k ,
|x|
then
∆Rn+1 u = 0.
Proof. Let M ⊂ Rn+1 be a hypersurface, u ∈ C 2 (Ω) and M ⊂ Ω ⊂ Rn+1
open. Let
∆ = ∆M ∧ ∆ ¯ = ∆Rn+1
and
(xα ), (ξ i )
coordinates for the ambient space and the hypersurface respectively. Then
we have
uij = uαβ xαi xβj + uα xαij
= uαβ xαi xβj − hij uα ν α .

⇒ ∆u = g ij uij = uαβ xαi xβj g ij − Huα ν α .

Choose, in a given point, coordinates such that
gij = δij ,
such that in this point we have
uαβ xαi xβj g ij = uαβ δiα δjβ g ij
= ḡ αβ uαβ − u00
= ḡ αβ uαβ − uαβ ν α ν β .

42
 ¯ − uαβ ν α ν β − Huα ν α .
⇒ ∆u = ∆u
Set λ = k(k + 1 − 1). On M = Sn we have H = n. Let u be homogeneous of
degree k in a neighborhood of Sn , then
 
k x
u(x) = |x| u .
|x|

Let (xα ) be euclidian coordinates, then

uα ν α = uα xα = ku.

⇒ kuβ xβ = uαβ xα xβ + uβ xβ
⇒ k(k − 1)u = (k − 1)uβ xβ = uαβ xα xβ

¯ + k(k − 1)u + nku

−∆u = −∆u
¯ + k(k + n − 1)u
= −∆u

1.7 The Harnack inequality

1.7.1 Assumptions of this section. Let Ω b Rn be open, n ≥ 2. In this
section we investigate the linear divergence form equation

Lu = −(aij uj )i + bi ui + cu = 0,

where
aij , bi , c ∈ L∞ (Ω),
kaij k∞ + kbi k∞ + kck∞ ≤ M
and
∃λ > 0 ∀ξ ∈ Rn : aij ξi ξj ≥ λ|ξ|2 .
1,2
1.7.2 Theorem. Let 0 ≤ u ∈ Hloc (Ω) and Lu ≤ 0, then for all B2R (x0 ) ⊂
Ω0 , q > 1, we have

ˆ !1
q
1 q
sup u ≤ c u ,
BR (x0 ) Rn B2R (x0 )

where c = c(Ω0 , n, q, λ, M ).

43
Proof. In this proof we use the so called Moser iteration technique.
(1) Suppose first that
u ∈ L∞ (B2R (x0 )).
Let p > 1,
η ∈ Cc0,1 (B2R (x0 )), 0 ≤ η ≤ 1,
uδ = u + δ and use up−1 2
δ η as a test function. Then
ˆ ˆ
2 p−2 2 p
(p − 1) |Du| uδ η ≤ c |Du||Dη|u−1 δ η(uδ dx)
Ω ˆΩ ˆ
+ c |Du|uδ (η uδ dx) + c upδ η 2
−1 2 p

ˆ
Ω
ˆΩ
c 2 p c
≤ |Du|2 u−2
δ η uδ + 2 |Dη|2 upδ
2 Ω
ˆ ˆΩ ˆ
c 2 −2 2 p c
+ |Du| uδ η uδ + η 2 upδ + c upδ η 2 .
2 Ω 2 Ω Ω

p−1
Setting  = 2c implies
ˆ ˆ
c
(p − 1) |Du|2 up−2 η 2 ≤ (|Dη|2 + η 2 )upδ . (1.18)
Ω p−1 Ω
Set
v = upδ η 2 .

ˆ ˆ ∞ ˆ
|Dv| ≤ p |Du||uδp−1 |η 2 + 2 upδ |Dη|η
Ω −∞ Ω
ˆ ˆ
p2 1
≤ (p − 1) |Du|2 uδp−2 η 2 + upδ η 2
p − 1 4
ˆ Ω Ω
p
+ 2 uδ |Dη|η.
Ω

Setting  = R and observing

n
H 1,1 (Ω) ,→ L n−1 (Ω)

we conclude
ˆ  n−1 ˆ
p2

p n n n 1 2 p
uδ n−1 η 2 n−1 ≤c +1 (R|Dη|2 + η 2 R + η )uδ .
Ω p−1 Ω R
n
For r ∈ N we set p = qκr , κ = n−1 and

R
ρr = R + .
2r

44
Choose (
1, x ∈ Bρr+1
η=
0, x ∈
/ Bρr ,
such that
1 2r+1
|Dη| ≤ = .
ρr − ρr+1 R
ˆ !1
κ ˆ
r+1
r 1 r
⇒ uqκ
δ ≤ c8 uqκ
δ
Bρr+1 R Bρr
ˆ !1
κ ˆ
1 qκr+1 1 r
⇒ uδ ≤ c8 n r
uqκ
δ
Rn Bρr+1 R Bρr

ˆ ! 1
κr+1
ˆ ! 1
κr
1 qκr+1 1 r 1 qκr
⇒ uδ ≤c κr 8 κr uδ .
Rn Bρr+1 Rn Bρr

This inequality is of the form

1 r
∀r ∈ N : Ir+1 ≤ c κr 8 κr Ir ,

which implies Pr Pr
1 i
Ir+1 ≤ c i=0 κi 8 i=0 κi I0
and thus ˆ
1
sup uqδ ≤c n uqδ .
BR R B2R

δ → 0 implies the claim.

(2) We now prove that u ∈ L∞

loc (Ω).
Define
∀1 ≤ p < ∞ : v = log(u + 1) ∈ Lploc (Ω).
Let p ≥ 2, then for η ∈ Cc0,1 (Ω) we have the test function

v p−1 η 2 ∈ H01,2 (Ω).

ˆ ˆ ˆ
p−2 2 p−1
(p − 1) Du · Dvv η ≤c |Du||Dη|v η+c |Du|v p−1 η 2
Ω ˆΩ Ω
p−1 2
+c v η u
Ω

As in (1.18) we obtain
ˆ ˆ
⇒ (p − 1) |Dv|2 v p−2 η 2 (u + 1) ≤ c (|Dη|2 + η 2 )(v p−1 + v p )(u + 1).
Ω Ω

45
 n
H 1,1 (Ω) ,→ L n−1 (Ω) implies
ˆ  n−1 ˆ
n n
p−1 2
(v η (1 + u)) n−1 ≤ c(p − 1) |Dv|v p−2 η 2 (1 + u)
Ω
ˆ Ω
p−1 (1.19)
+ c v |Dη|η(u + 1)
ˆΩ
+ c v p−1 η 2 |Dv|(u + 1).
Ω

Thus
ˆ  n−1 ˆ
n n
p−1 2
(v η (1 + u)) n−1 ≤ c(p−1) (|Dη|2 +η 2 )(v p−2 +v p−1 +v p )(u+1).
Ω Ω

Note that n
v p (u + 1) ≤ cv p−1 (u + 1) n−1
and
v p−2 ≤ v p−1 + 1,
since p ≥ 2. Thus
ˆ  n−1 ˆ
n n n
p−1 2
(v η (1 + u)) n−1 ≤ c(p − 1) (|Dη|2 + η 2 )(1 + v p−1 )(u + 1) n−1 .
Ω Ω
R n
Choose η as in part (1), ρr = R + 2r , κ= n−1 .

ˆ !1
κ ˆ
(p−1)κ κ 1 r
⇒ v (1 + u) ≤ c(p − 1)8 2 (1 + v p−1 )(1 + u)κ .
Bρr+1 R Bρr

There holds
ˆ !1
κ ˆ !1
κ

R−n (v (p−1)κ + 1)(1 + u)κ ≤ R−n v (p−1)κ (1 + u)κ

Bρr+1 Bρr+1
ˆ !1
κ

+ R−n (1 + u)κ .
Bρr+1

Then
ˆ !1
κ ˆ !
1 1
(v (p−1)κ + 1)(1 + u)κ ≤ c(p−1)8r (1 + v p−1 )(1 + u)κ .
Rn Bρr+1 Rn+1 Bρr

Set p − 1 = κr . For
ˆ ! 1
κr
1 κr κ
Ir = (v + 1)(1 + u)
Rn Bρr

46
we find c 1
κr r r
Ir+1 ≤ κ κr 8 κr Ir .
R
n
As is part (1) we conclude, using (1.19) with p = n−1 + 1,
ˆ
1 1
sup v ≤ c n ((v + 1) κ (1 + u))κ
BR R B2R
ˆ
1
≤c n (1 + u)p
R B2R
1
≤ c kuk1,2,B2R .
R

1,2
1.7.3 Theorem. Let 0 ≤ u ∈ Hloc (Ω) and Lu ≥ 0. Then for all B2R ⊂ Ω
and for all q < 0 we have
 ˆ 1
1 q
q
inf u ≥ c u ,
BR Rn B2R

where c = c(L, q, n).

Proof. Let δ > 0, uδ = u + δ and p < 1. Let 0 ≤ η ∈ Cc0,1 (B2R ) and multiply
the inequality by
up−1
δ η .
2

As in the previous theorem we conclude

ˆ ˆ 
|Dη|2

2 p−2 2 c
|p − 1| |Duδ | uδ η ≤ + η upδ .
2
Ω |p − 1| Ω |p − 1|

As in the proof of the previous theorem we obtain, using the -trick, that
ˆ  n−1 ˆ 
p2
 
p n n n 1 2
uδ n−1 η 2 n−1 ≤c +1 η upδ . R|Dη|2 + η 2 R +
Ω |p − 1|
Ω R
(1.20)
n r
Choose q < 0, κ = n−1 , p = qκ , r ∈ N. Using Moser iteration we obtain
 ˆ 
1
sup uqδ ≤c uqδ
BR Rn B2R

and since q < 0 we have

 ˆ 1
1 q
inf uδ ≥ c uqδ .
BR Rn B2R

For δ → 0 we obtain the claim.

47
 1,2
1.7.4 Lemma. Let 0 ≤ u ∈ Hloc (Ω) and Lu = 0. Let B4R ⊂ Ω. Then for
all 0 < q < 1 with the property
 r
n
∀r ∈ N : q 6 1
=
n−1
we have  ˆ 1
1 q
q
sup u ≤ c u ,
BR Rn B4R

where c = c(L, n, q).

n
Proof. Set κ = n−1 . Let r0 be minimal, such that

qκr0 > 1, R̃ = 2R.

Let
R̃
ρr = R̃ +
2r
and
p = qκr , 0 ≤ r ≤ r0 − 1.
Let η be as in the proof of 1.7.2. Using (1.20) we obtain, using R̃ instead of
R, as well as Lu ≥ 0,
ˆ ! 1
κr
1 qκr
Ir+1 ≤ cIr = c u .
Rn Bρr

Thus  ˆ  1  ˆ 1
1 qκr0 1 q
qκr0 q
u ≤c u .
Rn B2R Rn B4R

By 1.7.2 We obtain, using Lu ≤ 0,

 ˆ 1
1 q
q
sup u ≤ c u .
BR Rn B4R

1,2
1.7.5 Corollary. Let 0 ≤ u ∈ Hloc (Ω), Lu = 0 and B4R ⊂ Ω. Then for all
0 < q ∈ R we have
 ˆ 1
1 q
q
sup u ≤ c n
u ,
BR R B4R
where c = c(L, n, q).

Proof. Since the estimate holds for all q > 1 and for a dense subset of
{0 < q < 1}, we obtain the claim using the Hölder inequality.

48
1.7.6 Theorem. Let B = BR and suppose that u ∈ H 1,1 (B) satisfies
ˆ
∀Bρ (x0 ), x0 ∈ B, 0 < ρ < 2R : |Du| ≤ Aρn−1 . (1.21)
B∩Bρ (x0 )

Then there exists c = c(n), such that

ˆ
1
∀0 < b ≤ : eb|u−uB | ≤ c|B|,
cA B

1
´
where uB = |B| B u.

Proof. Let u ∈ Cc1 (Rn ), x, y ∈ B. Without loss of generality suppose x = 0

and choose polar coordinates around x to obtain
ˆ |y|
u(x) − u(y) = − ur dr.
0

ˆ ˆ ˆ |y|
1 c
|u(x) − u| ≤ n |Du(r, ξ)|χB drdy
|B| B R B2R (x0 ) 0
ˆ ˆ 2R ˆ 2R
−n n−1
≤ cR t |Du(r, ξ)|χB drdtdHn−1
Sn−1 0 0
ˆ ˆ 2R ˆ 2R
−n n−1 |Du(r, ξ)|
= cR t rn−1 χB
Sn−1 0 0 rn−1
ˆ 2R ˆ
|Du(y)|
= cR−n tn−1
0 B∩B2R (x0 ) |x − y|n−1
ˆ
|Du(y)|
=c .
B |x − y|n−1

Thus we have
ˆ ˆ ˆ p
1,1 p p |Du(y)|
∀u ∈ H (B) : |u − uB | ≤ c . (1.22)
B B B |x − y|n−1

We have
1 1
|Du(y)| |Du(y)| p |Du(y)| p0
= n−1 n−1 , p ≥ 2.
|x − y|n−1 +1 −1
|x − y| p 2p |x − y| p0 2p

Thus
ˆ ˆ ˆ ! ˆ !p−1
|Du| |Du|
|u − uB |p ≤ cp 1 1
B B B |x − y|n−1+ 2 B |x − y|
n−1− 2(p−1)

(1.23)

49
Set Du(y) = 0 for y 6= B define for x ∈ B, α > 0
ˆ
|Du(y)|
Iα (u) =
|x − y|n−1−α
ˆB
|Du(y)|
= n−1−α
|x−y|<2R |x − y|
∞ ˆ
X |Du(y)|
=
R R |x − y|n−1−α
t=0 2t <|x−y|< 2t−1
X∞ ˆ
≤ (2t R−1 )n−1−α |Du(y)|
t=0 |x−y|<21−t R
X∞
≤A (2t R−1 )n−1−α (21−t R)n−1
t=0
∞
X
= AR 2 α n−1
2−αt
t=0
1
= ARα 2n−1 .
1 − 2−α
The last integral in (1.23) is I 1 (u), p ≥ 2. There holds
2(p−1)

1
∀p ≥ 2 : 1 ≤ c0 p,
− 2(p−1)
1−2
because:
1
1
Set t = p−1 . We have 1 − 2− 2 t = 1 − e−at , a > 0. Since

1 − e−at
→ a,
t
we have
a
1 − e−at ≥ t,
2
from which the claim follows. Thus

I 1 ≤ ARα 2n−1 c0 p
2(p−1)

In (1.23) this reads

ˆ ˆ ˆ !
p−1 p−1 12 dx
p
|u − uB | ≤ cp cp−1
0 p A R |Du(y)| 1 .
B B B |x − y|n−1+ 2
ˆ
∀p ≥ 2 : |u − uB |p ≤ c1 Rn (cc0 Ap)p . (1.24)
B

50
Using the potential estimates, (1.21), (1.22) to handle the case p = 1 and
(1.24) that
ˆ ∞ ˆ
X bp
eb|u−uB | = |u − uB |p
B B p!
p=0
∞
X (bcc0 Ap)p
≤ c1 Rn + |B|
p!
p=1

Let bcc0 A ≤ κe , then the series converges by the quotient criterion and
ˆ
∀0 < b ≤ b0 : eb|u−uB | ≤ cRn .
B

1,2
1.7.7 Lemma. Let 0 ≤ u ∈ Hloc (Ω), Lu ≥ 0 and v = log u. Then there
holds for all B2ρ ⊂ Ω and ρ < 1, that
ˆ
|Dv| ≤ Aρn−1 .
Bρ

Proof. Let  > 0, v = log(u + ), 0 ≤ η ∈ C00,1 (B2ρ ) such that

1
η|Bρ = 1 ∧ |Dη| ≤ .
ρ

Multiply Lu ≥ 0 by (u + )−1 η 2 and set u = u + .

ˆ ˆ
|Du | u η ≤ c |Du |u−1
2 −2 2
 |Dη|η
Ω ˆ Ω ˆ
+ c |Du |u−1 η 2
+ c η2
Ω Ω
ˆ ˆ ˆ
1
⇒ |Dv |2 ≤ c +c 1 ≤ cρn−2 .
Bρ B2ρ ρ2 B2ρ

ˆ ˆ !1
2
n
⇒ |Dv | ≤ c |Dv |2 ρ 2 ≤ cρn−1 .
Bρ Bρ

For  → 0 we obtain the claim.

1,2
1.7.8 Lemma. Let 0 ≤ u ∈ Hloc (Ω) and Lu ≥ 0. Then there exist α, c > 0
such that for all B2ρ ⊂ Ω and ρ < 1

ˆ !1
α ˆ !− 1
α
1 1
|u|α ≤c |u|−α .
ρn Bρ ρn Bρ

51
Proof. Set v = log u. Then by 1.7.6 and 1.7.7 we have
ˆ
eb|v−vB | ≤ cρn
Bρ

for small b. Thus

ˆ ˆ
b(v−vB )
e ≤ cρ n
∧ e−b(v−vB ) ≤ cρn .
Bρ Bρ

Multiplying those inequalities we obtain

ˆ ˆ !−1
1 b 1 −b
u ≤c u .
ρn Bρ ρn Bρ

1.7.9 Theorem. (Weak Harnack inequality)

1,2
Let 0 ≤ u ∈ Hloc (Ω) and Lu ≥ 0. Let B4ρ ⊂ Ω, ρ < 1. Then there is p > 0,
such that
ˆ !1
p
1 p
u ≤ c inf u,
ρn Bρ Bρ

where c = c(n, L, p).

Furthermore there holds
1,2
1.7.10 Theorem. Let 0 ≤ u ∈ Hloc (Ω), Lu = 0. Then for all B4ρ ⊂ Ω and
0 < ρ < 1, there holds

sup u ≤ c inf u, c = c(n, L)

Bρ Bρ

and for connected Ω0 b Ω we have

sup u ≤ c inf0 u, c = c(n, L, Ω0 ).

Ω0 Ω

1,2
1.7.11 Theorem. Let 0 ≤ u ∈ Hloc (Ω), Lu ≥ 0 and Ω connected. If for
B ⊂ Ω we have inf B u = 0, then u ≡ 0 in Ω.
Proof. Follows immediately from the previous theorems.

1.7.12 Theorem. (Strong maximum principle)

1,2
Let Ω be connected and u ∈ Hloc (Ω), Lu ≤ 0 and c ≥ 0. If for a ball B ⊂ Ω
we have supB u = supΩ u ≥ 0, then u ≡ const.
Proof. Set M = supΩ u ≥ 0, v = M − u ≥ 0. Then Lv ≥ 0, since c ≥ 0.
Then the previous theorem implies the claim.

52
 Chapter 2

Hölder continuity of weak

solutions

2.1 Solution of the homogeneous equation

2.1.1 Lemma. Let ω ∈ L∞
loc (0, ρ0 ) suffice

ω(ρ) ≤ aω(4ρ) + kρα

for 0 < 4ρ < ρ0 < 1, 0 < a < 1, k ≥ 0, 0 < α < 1. Then we have

∀0 < R < ρ0 ∃c > 0 ∃0 < λ ≤ α ∀0 ≤ ρ ≤ R : ω(ρ) ≤ cρλ , λ = λ(a, α).

Proof. Choose 0 < β < 1 and a0 , such that a4β = a0 < 1. Set λ = min(α, β).
Let R4 ≤ ρ < R and
ω(ρ)
M = sup λ
.
R
≤ρ<R ρ 4

Then
R
∀ ≤ ρ < R : ω(ρ) ≤ M ρλ .
4
R R
Let 42
≤ρ< 4. Then

⇒ ω(ρ) ≤ aω(4ρ) + kρλ

≤ aM (4ρ)λ + kρλ
= (aM 4λ + k)ρλ
R R
By induction we then have in 4i+1
≤ρ< 4i

i
X
λ i
ω(ρ) ≤ (M (4 a) + k 4jλ aj )ρλ ,
j=0

53
 R R
since it holds for i = 0 and if it holds for i − 1, then for 4i+1
≤ρ< 4i
we
have

ω(ρ) ≤ aω(4ρ) + kρλ

i−1
X
≤ a(M (4λ a)i−1 + k 4jλ aj )(4ρ)λ + kρλ
j=0

i
X
ω(ρ) ≤ (M (4λ a)i + k 4jλ aj )ρλ
j=0
∞
aj0 )ρλ
X
≤ (M + k
j=0
1
= (M + k )ρλ ∀0 < ρ < R,
1 − a0
R R
since every ρ lies in a 4i
≤ρ< 4i−1
.
1,2
2.1.2 Theorem. Let Ω b Rn be open, n ≥ 2, and let u ∈ Hloc (Ω) be a
solution of the equation

Lu = −(aij uj )i + bi ui = 0,

where aij , bi ∈ L∞ (Ω) and aij is uniformly elliptic. Then u ∈ C 0,α (Ω),
α = α(n, L).

Proof. By the previous lemma it suffices to derive an estimate for the oscil-
lation of u,
ωρ = sup u − inf u,
Bρ Bρ

for every ball B4ρ (x) ⊂ Ω, such that

∀x ∈ Ω ∃0 ≤ a < 1 ∀ρ ≤ 1 : ωρ ≤ aω4ρ .

So let B4ρ b Ω and define

m(ρ) = inf u ∧ M (ρ) = sup u.

Bρ Bρ

⇒ v = M (4ρ) − u ≥ 0 in B4ρ .
Thus v is a nonnegative solution and by the Harnack inequality we obtain

sup v = M (4ρ) − m(ρ) ≤ c inf v = c(M (4ρ) − M (ρ)).

Bρ Bρ

54
Similarly for w = u − m(4ρ) ≥ 0 we find

sup w = M (ρ) − m(4ρ) ≤ c(m(ρ) − m(4ρ))

Bρ

and thus
ω4ρ + ωρ ≤ c(ω4ρ − ωρ )
c−1
⇒ ωρ ≤ ω4ρ .
c+1

2.2 Local Hoelder continuity

2.2.1 Assumptions of this section. Let Ω b Rn be open, n ≥ 2. We
1,2
consider solutions u ∈ Hloc (Ω) of

Lu = −(aij uj )i + bi ui + cu = −(f i )i ,

where
aij , bi , c ∈ L∞ (Ω),
(f i ) ∈ Lp (Ω, Rn ), n < p < ∞
and aij is uniformly elliptic. Furthermore we define the operator

L̃ = L − c.
1,2
2.2.2 Theorem. Let u ∈ Hloc (Ω) be a solution of Lu = −(f i )i . Then u is
locally bounded.

Proof. Will be proven more generally in a later theorem.

2.2.3 Lemma. (Stampacchia)

Let 0 ≤ φ : [k0 , ∞) → R be a nonincreasing function satisfying
c
∀h > k ≥ k0 : φ(h) ≤ φ(k)β (2.1)
(h − k)α

with positive constants α, β, c, then there hold

αβ
(1) β > 1 ⇒ φ(k0 + d) = 0, where dα = cφ(k0 )β−1 · 2 β−1 ,
1
(2) β = 1 ⇒ ∀h > k0 : φ(h) ≤ eφ(k0 )e−a(h−k0 ) , where a = (ec)− α and
µ 1
(3) β < 1 ∧ k0 ≥ 0 ⇒ φ(h) ≤ 2 1−β (c 1−β +(2k0 )µ φ(k0 ))h−µ , where µ = α
1−β .

Proof. (1) Consider the sequence

ki = k0 + d − d2−i , i ∈ N.

55
By (2.1) we obtain
c2α(i+1)
φ(ki+1 ) ≤ φ(ki )β (2.2)
dα
φ(k0 ) α
⇒ φ(ki ) ≤ iµ , µ = , (2.3)
2 β−1
since it holds for i = 0 and
c2α(i+1) φ(k0 )β φ(k0 )
φ(ki+1 ) ≤ = (i+1)µ .
dα 2iµβ 2
(2) Consider
1
ki = k0 + i(ec) α .
By (2.1) we have
1
φ(ki ) ≤ φ(ki−1 ).
e
Let h > k0 . Then there exists an i ∈ N, such that
1 1
k0 + (i − 1)(ec) α ≤ h ≤ k0 + i(ec) α .
1 1
φ(h) ≤ φ(k0 + (i − 1)(ec) α ) ≤ e−(i−1) φ(k0 ) ≤ ee−a(h−k0 ) φ(k0 ), a = (ec)− α .
(3) Let
hµ
ψ(h) = φ(h) 1 .
c 1−β
By (2.1) we have for all h > k ≥ k0 ≥ 0
β
hµ c c 1−β
ψ(h) ≤ 1 α µβ
ψ(k)β
(h
c 1−β  − k) k
µ
β h
= ψ(k) .
(h − k)1−β k β
h := 2k implies
ψ(2k) ≤ 2µ ψ(k)β (2.4)
Pi−1
βi βj
ψ(2i k) ≤ ψ(k) 2µ j=0 , (2.5)
since it holds for i = 1 and we have

ψ(2i+1 ) ≤ 2µ ψ(2i k)β

Pi−1
i+1 β j+1
≤ ψ(k)β 2µ j=0 · 2µ
Pi
i+1 βj
= ψ(k)β 2µ j=0 .

β < 1 implies
i 1
− 1−β
sup ψ(k)β ≤ 1 + sup ψ(k) ≤ 1 + φ(k0 )(2k0 )µ c
k0 ≤k≤2k0 k0 ≤k≤2k0

56
 1 1
− 1−β µ 1−β
⇒ ψ(2i k) ≤ (1 + φ(k0 )(2k0 )µ c )2 . (2.6)
Each number h ≥ 2k0 is of the form h = 2i k, k ∈ [k0 , 2k0 ]. Thus by (2.6) we
have
− 1 µ 1
sup ψ(h) ≤ (1 + φ(k0 )(2k0 )µ c 1−β )2 1−β
h≥k0
µ 1
⇒ φ(h) ≤ 2 1−β (c 1−β + φ(k0 )(2k0 )µ )h−µ .

2.2.4 Theorem. Let u ∈ H 1,2 (Ω) be a solution of

L̃u = −(f i )i ,

then there holds

(1) If bi = 0 or |Ω| ≤ 0 = 0 (n, L) << 1, then
1
− p1
|u| ≤ sup |u| + ckf kp |Ω| n .
∂Ω

(2) Otherwise there holds

1
− p1
|u| ≤ c0 + ckf kp |Ω| n ,

where c0 = c0 (n, sup∂Ω |u|, kuk1 ).

Proof. Let k ∈ R and set

A(k) = {u > k}.

Let k0 = max(sup∂Ω u, 0). For k ≥ k0 define

η = max(u − k, 0) ∈ H01,2 (Ω)

as test function. Then

ˆ ˆ ˆ
2
|Dη| ≤ c |f ||Dη| + c |Dη|η
Ω Ω Ω
ˆ ˆ ˆ
2 2
⇒ |Dη| ≤ c |f | + c |η|2
Ω A(k) A(k)

n ≥ 3 ⇒ 2∗ = 2n
n−2 .
ˆ ˆ
2
|η|2 ≤ c |Dη|2 |A(k)| n .
A(k) Ω

For n = 2 we have
ˆ ˆ 2 ˆ
2
|η| ≤ c |Dη| ≤c |Dη|2 |A(k)|.
A(k) Ω Ω

57
For small |Ω| we find
ˆ ˆ
2
∀k ≥ k0 : |Dη| ≤ c |f |2 .
Ω A(k)

If Ω is arbitrary, k0 has to be chosen large enough, depending on kuk1 and

n, since ˆ ˆ ˆ
u
|A(k)| = 1≤ ≤ k −1 |u|.
A(k) A(k) k Ω

Then ˆ ˆ
2 1− p2
∀k ≥ k0 : |Dη| ≤ c |f |2 ≤ ckf k2p |A(k)| .
Ω A(k)
ˆ  n−1 ˆ ˆ 1
n n 2 1
2
η n−1 ≤c |Dη| ≤ c |Dη| |A(k)| 2
Ω Ω Ω
ˆ  n−1
n
n 1− p1
⇒ η n−1 ≤ c|A(k)| kf kp .
Ω
ˆ ˆ  n−1
n 1
n 1 1+ n − p1
η≤ η n−1 |A(k)| n ≤ c|A(k)| kf kp .
Ω Ω
Now for all h > k ≥ k0 we have
ˆ ˆ
1+ 1 − 1
(h − k)|A(h)| ≤ (u − k) ≤ η ≤ ckf kp |A(k)| n p .
A(h) Ω

1 1
For β = 1 + n − p > 1 we have by 2.2.3

|A(k0 + d)| = 0,
1
− p1
where d = ckf kp |A(k0 )| n .

⇒ u ≤ k0 + d.

Analogously this holds for −u, which implies the claim.

1,2
2.2.5 Theorem. Let u ∈ Hloc (Ω) be a solution of

Lu = −(f i )i , f i ∈ Lploc (Ω), p > n,

then u ∈ C 0,α (Ω).

Proof. If u ∈ L∞
loc (Ω), we may consider

L̃u = −cu − fii ≡ g − fii .

Let Ω ≡ Ω0 b Ω, then we claim:

∃w ∈ C 0,1 (Rn ) : − ∆w = g = −(δ ij wj )i .

58
Proof : Extend g identically 0 to Rn and call the mollification g . Set

ω = γ ∗ g ,

where γ is the Newtonian potential. Then we have

−∆ω = g

and
|Dω | ≤ const.
As  → 0 we obtain a limit

ω → ω ∈ C 0,1 (Ω̄) : − ∆ω = g.

Thus without loss of generality we may assume g = 0.

Now let B4ρ b Ω and ρ so small that L̃ coercitive in H01,2 (B4ρ ), i.e.

∀u ∈ H01,2 (B4ρ ) : hL̃u, ui ≥ ckuk21,2 .

Then solve

L̃w = −(f i )i in B4ρ ≡ B

w|∂B4ρ = 0.

Therefore define
a(u, v) = hL̃u, vi
and
φ ∈ H01,2 (B)∗
by ˆ
v 7→ h−fii , vi = f i vi .
Ω

Then a induces a linear operator A ∈ L(H01,2 (B)).

Thus the above equation
reduces to
hAw, vi = hφ, vi ∀v ∈ H01,2 (B).
There exists a solution by Exercises 13. Thus by the previous theorem we
obtain for such a solution
1− n
|w| ≤ ckf kp ρ p .

Set v = u − w, then
L̃v = 0 in B4ρ .
Let ωv = osc(v), then as in the proof of theorem 2.1.1 we obtain

ωv (ρ) ≤ aωv (4ρ), 0 < a < 1.

59
 ωu (ρ) ≤ ωv (ρ) + ωw (ρ)
and
ωv (4ρ) ≤ ωu (4ρ) + ωw (4ρ)
1− n
⇒ ωu (ρ) ≤ aωu (4ρ) + ckf kp ρ p
n
⇒ ∃0 < α ≤ 1 − ∀0 < ρ̃ ≤ 2ρ : ωu ≤ cρ̃α .
p

2.2.6 Lemma. (Stampacchia)

Let 0 ≤ φ(k, ρ) be a real function, k > k0 , 0 < ρ < R0 such that
(1) φ(·, ρ) is monotonely decreasing,
(2) φ(k, ·) is monotonely increasing and
c 1 β
(3) ∀k0 < k < h ∀0 < ρ < R < R0 : φ(h, ρ) ≤ (h−k) α (R−ρ)γ |φ(k, R)| ,
c, α, β, γ > 0, β > 1.
Then there holds

∀0 < σ < 1 : φ(k0 + d, R0 (1 − σ)) = 0

with
β
(α+γ) β−1 |φ(k0 , R0 )|β−1
dα = 2 c .
σ γ R0γ
d σR0
Proof. Consider ki = k0 + d − 2i
, ρi = R0 − σR0 + 2i
. Then there holds

φ(k0 , R0 ) α+γ
φ(ki , ρi ) ≤ µi
, µ= ,
2 β−1
since it clearly holds for i = 0 and

φ(ki+1 , ρi+1 ) ≤ cφ(k0 , R0 )β 2−µβi d−α · 2(i+1)(α+γ) σ −γ R0−γ

φ(k0 , R0 )
= µ(i+1) .
2

2.2.7 Theorem. Let Ω b Rn be open and ∂Ω ∈ C 0,1 . Let u ∈ H 1,2 (Ω) be a

weak solution of
Lu = −fii , f i ∈ Lp (Ω), p > n.
Then there hold
(1) supBρ |u| ≤ c(kuk2,B2ρ , kf kp , n, p, L) ∀B2ρ b Ω and
(2) Let x0 ∈ ∂Ω, Ωρ (x0 ) = Ω ∩ Bρ (x0 ), Γρ = ∂Ω ∩ Bρ (x0 ) and suppose
supΓ2ρ |u| ≤ γ < ∞, then there holds

sup |u| ≤ c(γ, kf kp , kuk2,Ω2ρ , L, p, n).

Ωρ

60
Proof. We only prove part (2), since the first part works identically. Let
0 < ρ1 < ρ2 < 2ρ < 1 and

0 ≤ η ∈ C00,1 (Bρ2 (x0 )),

such that
1
η|Bρ1 = 1 ∧ |Dη| ≤ .
ρ2 − ρ1
Furthermore let k0 ≥ max(γ, 1), v = log u on {u > 0} and

vk = max(v − k, 0), k ≥ k0 .

Thus, if vk > 0, it follows u > 1. Multiply the equation by

vk η 2 ∈ H01,2 .

Then, using the -trick,

ˆ ˆ ˆ
|Dvk | η u ≤ c |b||Dvk |vk uη + c vk η 2 u
2 2 2
Ω ˆΩ ˆ Ω
+ c |f ||Dvk |η 2 + c |f |vk |Dη|η
ˆΩ Ω
2 2
+ c |vk | |Dη| u.
Ω

Define
A(k, η) = {vk η 2 > 0}, A(k, ρ) = {v > k} ∩ Bρ (x0 )
as well as the measure ˆ
|A(k, η)| = u.
A(k,η)

ˆ ˆ
2 2
|Dvk | η u ≤ c vk2 (η 2 + |Dη|2 )u
Ω ˆΩ ˆ
2 −1 2
+c |f | u |η| + c vk η 2 u.
A(k,η) Ω

ˆ 1 ˆ !1
p
2 c  1
− p1
2 2
|Dvk | η u ≤ vkp u |A(k, η)| 2
Ω ρ2 − ρ1 A(k,η)
ˆ !1
p
1
p − p1
+ |f | |A(k, η)| 2
A(k,η)
ˆ 1
2r 1 1

+ vkr u |A(k, η)| 2 − 2r
Ω

61
 n
Setting κ = n−1 and applying the Sobolev embeddings we obtain
ˆ 1 ˆ
κ
2 κ
(vk η u) ≤c (|Dvk |η 2 u + vk |Dη|ηu + uvk η 2 |Dvk |)
Ω Ω
ˆ 1
2 1
2 2
≤c |Dvk | η u |A(k, η)| 2
Ω
ˆ 1
1 r
r 1
+c vk u |A(k, η)|1− r
ρ2 − ρ1 Ω
ˆ  1 ˆ 1
2 2
2 2 2 2
+c |Dvk | η u vk η u .
Ω Ω

Thus
ˆ ˆ !1 

1 p
κ c  |A(k, η)|1− p
1
(vk η 2 u)κ ≤ kf kp + vkp u
Ω ρ2 − ρ1 A(k,η)

ˆ !1
r
c 1
+ vkr u |A(k, η)|1− r
ρ2 − ρ1 A(k,η)
ˆ ! 1  ˆ

p 1
c r
1− r1 − p1
+ kf kp + vkp u  vkr uη 2 |A(k, η)| .
ρ2 − ρ1 A(k,η) Ω

(2.7)

Since ∀1 < r < ∞ we have

ˆ !1
r ˆ !1
2

vkr u ≤ cr |u| 2
,
A(k,η) A(k,η)

it follows ˆ 1
2 κ
κ c̃ 1− 1 − 1
(vk η u) ≤ |A(k, η)| r p .
Ω ρ2 − ρ1
ˆ
c̃ 1+ 1 − 1 − 1
⇒ (h − k)|A(k, η)| ≤ vk η 2 u ≤ |A(k, η)| n r p .
Ω ρ2 − ρ1
1 1 1 1 1 1
Choose r such that r < n − p and set β = 1 + n − r − p > 1. Then for
h > k > k0 we find
c̃ 1
|A(h, ρ1 )| ≤ |A(k, ρ2 )|β ∀0 < ρ1 < ρ2 < 1.
ρ2 − ρ1 h − k
Then by 2.2.6 we obtain
|A(k0 + d, ρ)| = 0

62
with
β |A(k0 , 2ρ)|β−1
d = 4 β−1 c̃
ρ
⇒ sup u ≤ k0 + d.
Bρ

The same for −u implies the claim.

2.3 Hoelder estimates near the boundary

2.3.1 Assumptions of this section. Let Ω b Rn be open, n ≥ 2. We
consider solutions u ∈ H 1,2 (Ω) of

Lu = −(aij uj )i + bi ui + cu = −(f i )i ,
u|∂Ω = φ,

where
aij , bi , c ∈ L∞ (Ω),
(f i ) ∈ Lp (Ω, Rn ), n < p < ∞,
φ ∈ C 0,α (∂Ω), 0 < α ≤ 1 and aij is uniformly elliptic. Furthermore we define
the operator
L̃ = L − c.
2.3.2 Definition. We say, ∂Ω satisfies an exterior cone condition, ∂Ω ∈ (K),
if for each x0 ∈ ∂Ω there is a cone with uniform angle starting in x0 , such
that for a uniform ρ > 0 we have

Kρ (x0 ) = K ∩ Bρ (x0 ) ⊂ Ωc .

2.3.3 Example. ∂Ω ∈ C 0,1 ⇒ ∂Ω ∈ (K).

|Bρ (x0 )\Ω|
2.3.4 Remark. ∂Ω ∈ (K) ⇒ ∃0 > 0 ∀x0 ∈ ∂Ω : ρn ≥ 0 .
2.3.5 Theorem. Let 0 ≤ u, Lu ≥ 0, x0 ∈ ∂Ω and R > 0. Set

m = inf{u(x) : x ∈ ∂Ω ∩ B4R (x0 )}

and (
min(u, m), x ∈ Ω ∩ B4R
ū =
m, x ∈ B4R \Ω.
Then there holds for all p < 0
 ˆ 1
1 p
p
ū ≤ c inf ū,
Rn B2R BR

c = c(n, L, p).

63
Proof. Let p < 1, η ∈ C00,1 (B2R ), δ > 0, ūδ = ū+δ and mδ = m+δ. Multiply
Lu ≥ 0 by the test function

(ūp−1
δ − mδp−1 )η 2 ∈ H01,2 (Ω).

As in the proof of 1.7.3 we obtain

ˆ ˆ 
|Dη|2

2 p−2 2
|p − 1| |Dūδ | ūδ η ≤ c + η ūpδ .
2
Ω Ω |p − 1|

We also may integrate outside Ω in the full ball. Using the -trick and
n
Sobolevs embedding, κ = n−1 , R < 1, we obtain
ˆ 1 ˆ
p2

κ 1 2 p
ūκδ η 2κ ≤c +1 (R|Dη|2 + η 2 + η )ūδ . (2.8)
B2R |p − 1| B2R R

Let q < 0 and p = qκr , r ∈ N. Then by iteration we obtain

 ˆ 1
1 q
ūqδ ≤ c inf ūδ .
Rn B2R BR

δ → 0 implies the claim.

2.3.6 Lemma. Under the assumptions of the preceding theorem let 0 < q <
1. Then there is p > 1, such that
 ˆ 1  ˆ 1
1 p
p 1 q
q
ū ≤c ū ,
Rn B2R Rn B4R

c = c(n, L, p, q).

Proof. It suffices to prove the claim for almost every 0 < q < 1. So let
n
0 < q < 1 such that qκr 6= 1 for all r ∈ N, κ = n−1 . Choose r0 minimally
r
such that p = qκ > 1 then by (2.8) we obtain using iteration
0

 ˆ 1 ˆ
1 p
p 1 1
ū ≤ c( ūq ) q .
Rn B2R Rn B4R

2.3.7 Lemma. Under the assumptions of the preceding lemma let Ω4R =
Ω ∩ B4R (x0 ) and v = log(ū). Then
ˆ
∀B2ρ (y) ⊂ B4R (x0 ) : |Dv| ≤ Aρn−1 .
Bρ

64
Proof. Let 0 ≤ η ∈ C00,1 (B2ρ ), η|Bρ = 1, |Dη| ≤ ρ1 and  > 0. Let furthermore
ū = ū +  and v = log ū . Using the test function
((ū + )−1 − (m + )−1 )η 2
we obtain
ˆ ˆ
|Dū |2 ū−2
 η
2
≤c |Dū |ū−1
 |Dη|η
Ω ˆΩ
+c |Dū |ū−1
 η
2

ˆΩ
+c η2
Ω

and thus ˆ ˆ
2 2 −2
|Dv | η ≤ cρ η 2 ≤ cρn−2 .
Ω Ω
ˆ
⇒ |Dv | ≤ Aρn−1 .
Bρ

2.3.8 Lemma. Under the assumptions of the preceding lemma there exist
α > 0 and c > 0 such that
 ˆ 1  ˆ − 1
1 α
α 1 −α
α
ū ≤ c ū .
R n BR Rn BR
Proof. As 1.7.8.

2.3.9 Theorem. (Weak Harnack inequality)

Let 0 ≤ u, Lu ≥ 0 and m, ū as in the preceding theorem, then there exist
p > 1 and c = c(n, p, L), such that
 ˆ 1
1 p
p
ū ≤ c inf ū.
R n BR BR

Proof. (i) ∃q > 0 such that

 ˆ 1  ˆ − 1
1 q
q 1 −q
q
ū ≤ ū
Rn B2R Rn B2R
≤ c inf ū.
BR

(ii) Furthermore there is p > 1, such that

 ˆ 1  ˆ 1
1 p
p 1 q
q
ū ≤ c ū
Rn BR Rn B2R

65
2.3.10 Corollary. Under the assumptions of the preceding theorem there
holds ˆ
1
ū ≤ c inf ū.
Rn BR BR

2.3.11 Theorem. Let u ∈ H 1,2 (Ω) be a solution of the equation

L̃u = 0 in ΩR0 = Ω ∩ BR0 (x0 ),

x0 ∈ ∂Ω. Let Γ = BR0 ∩ ∂Ω ∈ C 0,1 and φ = u|Γ ∈ C 0,α , 0 < α < 1. Then
there exists 0 < a < 1, such that for 0 < ρ < R40 and for

ω(ρ) = sup |u(x) − u(y)|

x,y∈Ωρ (z0 )

and
ω̃(ρ) = sup |u(x) − u(y)|
∂Ω∩Bρ (z0 )

we have
ω(ρ) ≤ aω(4ρ) + ω̃(4ρ).

Proof. Let
M (ρ) = sup u, m(ρ) = inf u,
Ωρ Ωρ

M̃ (ρ) = sup u, m̃(ρ) = inf u.

∂Ω∩Bρ ∂Ω∩Bρ

(i) Consider v = M (4ρ) − u ≥ 0 in Ω4ρ , then we have

L̃v = 0.

Thus by the preceding corollary we have

ˆ
1
v̄ ≤ c inf v̄
ρn Bρ Bρ

⇒ ρ−n v̄|Bρ \Ω| ≤ c inf v̄ ≤ c inf v ≤ c(M (4ρ) − M (ρ)).

Ωρ Ωρ

Since ∂Ω ∈ (K) we have

M (4ρ) − M̃ (4ρ) ≤ c(M (4ρ) − M (ρ)).

(ii) Set v = u − m(4ρ) ≥ 0 in Ω4ρ . Then L̃v = 0. Thus we again have

ˆ
1
v̄ ≤ c inf v̄ ≤ c inf v ≤ c(m(ρ) − m(4ρ)).
ρn Bρ Bρ Ωρ

⇒ m̃(4ρ) − m(4ρ) ≤ c(m(ρ) − m(4ρ)).

66
(iii) Add the two inequalities to obtain
ω(4ρ) − ω̃(4ρ) ≤ c(ω(4ρ) − ω(ρ)), c > 1
c−1 1
⇒ ω(ρ) ≤ ω(4ρ) + ω̃(4ρ).
c c

2.3.12 Theorem. Let ∂Ω ∈ C 0,1 and u ∈ H 1,2 (Ω) be a solution of the

Dirichlet problem
Lu = −fii
u|∂Ω = φ,
where f i ∈ Lp (Ω), p > n.
Let x0 ∈ ∂Ω, Γ4R = ∂Ω ∩ B4R (x0 ) and φ ∈ C 0,α (Γ4R ), then there holds
u ∈ C 0,λ (Ω ∪ ΓR ),
0 < λ ≤ min(α, 1 − np ).
Proof. (i) By 2.2.7 we have u ∈ L∞ (Ω2R ). Solving
−∆w = −cu
we obtain
L̃u = −(f i + Di w)i ≡ −fii .
(ii) Having extended the data to B8ρ , solve
L̃w = −fii in B8ρ
w|∂B8ρ = 0,
for such small ρ < 1, that L̃ is coercitive.
1− n
⇒ sup |w| ≤ cρ p kf kp .
Setting
v = u − w,
we have
L̃v = 0
in Ω8ρ . Thus by the preceding theorem we have
ωv (ρ) ≤ aωv (4ρ) + ω̃v (4ρ).

⇒ ωv (ρ) ≤ aωu (4ρ) + aωw (4ρ) + ω̃u (4ρ) + ω̃w (4ρ)

1− n
≤ aωu (4ρ) + ckf kp ρ p + c[φ]α ρα
By the De Giorgi lemma we obtain
ωu (ρ) ≤ cρλ .

67
2.4 Application to nonlinear equations
Consider a general elliptic PDE of second order

F (·, u, Du, D2 u) = 0,
∂F
aij = > 0,
∂uij
where F is uniformly elliptic in compact subsets of the domain of definition
of F. If the regularity of the equations admits, we may differentiate for xk
to obtain
∂F ∂F ∂F
0 = aij ukij + ukj + uk + ,
∂pj ∂u ∂xk
which is a linear equation for v = uk . If it is a priori possible to obtain
C 3 estimates, we thus obtain v ∈ C 2,α by Schauder theory. Obtaining C 3
estimates is quite difficult in general. The results of Evans, Krylov for the
elliptic case and Krylov, Safonov for the parabolic case ensure C 2,α estimates
only knowing C 2 bounds and the concavity of F (·, u, Du, ·). We now turn
our attention to quasilinear equations.

2.4.1 Assumptions of this section. Let Ω b Rn be open. We consider

the quasilinear equation

Au = −(ai (·, u, Du))i = f

(2.9)
u|∂Ω = φ,

where
∂ai
aij =
∂pj
is locally uniformly elliptic, f ∈ Lp (Ω), p > n ≥ 2, ∂Ω ∈ C 2 and ai ∈
C 1 (Ω̄ × R × Rn ).

2.4.2 Theorem. (i) Let u ∈ C 0,1 (Ω̄) be a solution of (2.9), φ ∈ H 2,p (Ω),
f ∈ Lp (Ω). Then we have
u ∈ H 2,p (Ω).
(ii) Suppose furthermore f ∈ C 0,α (Ω̄), φ ∈ C 2,α (Ω̄) and ∂Ω ∈ C 2,α , then we
have
u ∈ C 2,α (Ω̄).

For the proof we first need several things.

2.4.3 Theorem. Under the assumptions of the preceding theorem, (i), we

have
u ∈ C 1,α (Ω̄),
for some 0 < α ≤ 1 − np .

68
Proof. (i) u ∈ C 0,1 (Ω̄) ⇒ ai (·, u, pj ) ≡ ai (·, pj ) and

Λ|ξ|2 ≥ aij ξi ξj ≥ λ|ξ|2 , λ > 0.

The L2 estimates imply u ∈ H 2,2 (Ω) and

kuk2,2 ≤ c(kf k2 + kuk2 ).

(ii) Let 1 ≤ k ≤ n, v = uk ∈ H 1,2 (Ω). Use ζk as test function to obtain

 i   i
ij ∂a ∂a
−(a vj )i − v − = fk = (δki f )i
∂u i ∂xk i

⇒ −(aij vj )i ≡ −fii ,
f i ∈ Lp (Ω). By the De Giorgi-Nash results we obtain v ∈ C 0,α (Ω) with
corresponding a priori esimates.
(iii) Boundary estimates. By local flattening we may assume the equation
reads

Au = f in Ω = B1+ (0)
u|Γ = φ,

where Γ = ∂Ω ∩ {xn = 0}.

Let 1 ≤ k ≤ n − 1 and v = uk . Then v solves the Dirichlet problem

−(aij vj )i = −fii in Ω
(2.10)
v|Γ = φk ∈ C 0,β ,

β = 1 − np . De Giorgi-Nash implies

+¯
v ∈ C 0,α (BR (0)),

0 < R < 1.
(iv) In order to prove
un ∈ C 0,α (B +
1 (0)),
2

we have to prove a so-called Morrey condition for Du. Let v be defined

+
as in (iii). Let 0 < R < 1 and ξ ∈ BR (0). Choose 0 < ρ < 1, such that
0,1
B2ρ (ξ) ⊂ BR (0) and let η ∈ C0 (B2ρ (ξ)), such that η|Bρ = 1 and |Dη| ≤ ρ1 .
Distinguish two cases:
(1) B2ρ (ξ) ∩ Γ 6= ∅. Then we choose ξ0 ∈ B2ρ (ξ) ∩ Γ and multiply (2.10) by

(v − φk (ξ0 ) − (φk − φk (ξ0 )))η 2 = (v − φk )η 2 ∈ H01,2 (Ω), Ω = B1+ (0).

+
(2) B2ρ(ξ) ⊂ BR (0), then we multiply (2.10) by

(v − v(ξ))η 2 .

69
In both cases integrate by parts. We only consider case (1).

⇒ |v − φk (ξ0 )| = |v − v(ξ0 )| ≤ c|x − ξ0 |α ≤ cρα

and
|φk − φk (ξ0 )| ≤ c|x − ξ0 |β ≤ cρβ ≤ cρα .
ˆ ˆ
aij vi vj η 2 ≤ aij vj φki η 2
Ω ˆ
Ω

− 2 aij vj (v − φk (ξ0 ) − (φk − φk (ξ0 )))ηi η

ˆ Ω
+ f i (vi − φki )η 2
ˆ
Ω

+ 2 f i (v − φk (ξ0 ) − (φk − φk (ξ0 )))ηi η

Ω

By the standard -trick we obtain

ˆ ˆ ˆ
2 2 2 −2
|Dv| ≤ c |D φ| + cρ (|v − v(ξ0 )|2 + |φk − φk (ξ0 )|2 )
Bρ (ξ)∩Ω B2ρ(ξ) ∩Ω B2ρ ∩Ω
ˆ
+c (|f |2 + |D2 φ|2 )
B2ρ ∩Ω
ˆ
+ cρ−1 |f |(|v − v(ξ0 )| + |φk − φk (ξ0 )|)
B2ρ ∩Ω
≡ I1 + I2 + I3 + I4

We have
n− 2n
I1 ≤ ckφk22,p ρ p ,
I2 ≤ c([v]2α + [Dφ]2α )ρn−2+2α ,
n− 2n
I3 ≤ c(kf k2p + kD2 φk2p )ρ p

and
n−1− n +α
I4 ≤ ckf kp ρ p .
ˆ
⇒ |Dv|2 ≤ cL2 ρn−2+2λ ,
Bρ ∩Ω

λ = min(α, 1 − np ) and L2 = kφk22,p + kf k2p + [v]2α + [Dφ]2α + kf kp . This is the

Morrey condition.
Now we show, that v = un satisfies a Morrey condition as well. We use the
equation:
∂ai ∂ai
−aij uij − i − ui = f.
∂x ∂u
ˆ
⇒ |unn |2 ≤ cL2 ρn−2+2λ .
Bρ ∩Ω

70
Using the following lemma we obtain

v ∈ C 0,λ (B +¯
R (0)).
4

2.4.4 Lemma. (Morrey)

+
Let Ω = BR (0) or Ω = BR (0) and suppose for u ∈ H 1,p (Ω) and 1 ≤ p ≤ n
there holds ˆ
|Du|p ≤ cLp ρn−p+pλ , λ > 0 (2.11)
Bρ (ξ)

for all 0 < ρ ≤ R

and for all ξ ∈ B R (0) or for all ξ ∈ B +¯
4 R (0) respectively.
4 4
Then
u ∈ C 0,λ (B R¯(0))
4

or
u ∈ C 0,λ (B +¯
R (0))
4

respectively and
[u]λ ≤ cL.
Proof. Prove only the case Ω = BR (0). Let u ∈ C 1 (Ω) and x, ξ ∈ B R (0).
4
Set
1 |x − ξ|
x̄ = (x + ξ), ρ = .
2 2
For y ∈ Bρ (x̄) we then have
ˆ 1 ˆ 1
d
u(y) − u(ξ) = u(ty + (1 − t)ξ) = ui (y i − ξ i ).
0 dt 0
ˆ ˆ 1ˆ
⇒ |Bρ (x̄)|−1 |u(y) − u(ξ)| ≤ 2ρ|Bρ (x̄)|−1 |Du(ty + (1 − t)ξ)|.
Bρ (x̄) 0 Bρ (x̄)

Transform
z = ty + (1 − t)ξ, z̄ = tx̄ + (1 − t)ξ
to obtain
ˆ 1ˆ ˆ 1 ˆ
|Du(ty + (1 − t)ξ)|dydt = t−n |Du(z)|dz
0 Bρ (x̄) 0 Btρ (z̄)
ˆ 1 ˆ
−n n p−1 1
≤c t (tρ) p ( |Du|p ) p
0 Btρ (z̄)
ˆ 1
n p−1 n
−1+λ
≤c t−n (tρ) p L(tρ) p
0
ˆ
⇒ |Bρ(x̄) |−1 |u(y) − u(ξ)| ≤ cLρλ
Bρ(x̄)

71
and analogously for x = ξ.
ˆ
−1
⇒ |u(x) − u(ξ)| ≤ |Bρ (x̄)| (|u(y) − u(x)| + |u(y) − u(ξ)|) ≤ cLρλ .
Bρ(x̄)

Now let ∂Ω ∈ C 2,α , φ ∈ C 2,α (Ω̄), f ∈ C 0,α (Ω̄) and u ∈ C 2,α (Ω̄) be a solution
of the problem (2.9). If we are able to prove C 0 and C 1 estimates, then
by De Giorgi-Nash we obtain C 0,α coefficients, bounded by |u|1,α . Schauder
theory then yields C 2,α estimates.

We now prove that Lipschitz solutions are already classical solutions.

2.4.5 Theorem. Let Ω b Rn be open and let ∂Ω ∈ C 2,α ,

ai , a ∈ C 1,α (Ω̄ × R × Rn ), φ ∈ C 2,α (Ω̄) and u ∈ C 0,1 (Ω̄) a solution of

Au + a(·, u, Du) = 0
(2.12)
u|∂Ω = φ,

then we have
u ∈ C 2,β (Ω̄),
for some 0 < β ≤ α.

Proof. (i) Let u0 ∈ C 0,1 (Ω̄) be a solution and let

1 + |u0 | + |Du0 | ≤ M.

Let θ = θ(t) be a real function

(
t, |t| ≤ M
θ(t) =
±(M + 1), |t| ≥ M + 1,

θ̇ ≥ 0.
(ii) Let w, g be real functions
(
1, 0 ≤ t ≤ 2M
w(t) =
0, t ≥ 3M

and (
0, 0≤t≤M
g(t) =
ct − k, t ≥ 2M,
such that g is convex.
(iii) Set
ãi (x, t, p) = ai (x, θ(t), p)w(|p|2 ) + kg 0 (|p|2 )pi ,

72
where k is large. Furthermore set

ã(x, t, p) = a(x, θ(t), p)w(|p|2 ).

There holds
|ã(x, t, p)| ≤ c(1 + |p|)
and ãij is uniformly positive definite. Thus the corresponding operator

Ãu + ã(·, u, Du)

is a uniformly elliptic differential operator. If γ > 0 is chosen large enough,

then
Φu := Ãu + ã(·, u, Du) + γ(u − u0 )
is coercitive i.e. for u1 , u2 ∈ H 1,2 (Ω) such that u1 = u2 on ∂Ω we have

hΦu1 − Φu2 , u1 − u2 i ≥ cku1 − u2 k21,2 , c > 0.

Using the exercises we obtain u ∈ H 1,2 (Ω), solving

Φu = 0
(2.13)
u|∂Ω = φ.

By L2 estimates and De Giorgi-Nash we obtain

u ∈ C 1,α (Ω̄) ∩ H 2,2 (Ω).

There holds
Φu0 = Au0 + a(·, u0 , Du0 ).
Thus, if (2.13) has a C 2,α solution u, then we must have u = u0 .
(iv) (2.13) has a C 2,β (Ω̄) solution. The linearization reads

−ãij uij + â(·, u, Du) + γ(u − u0 ) = 0

(2.14)
u|∂Ω = φ

with ∂â∂t + γ > 0.

First, we need an a priori estimate:
(1) By the maximum principle we obtain a C 0 estimate.
(2) Using Thm 15.2 in Gilbarg-Trudinger we obtain a C 1 estimate, also cf.
Chapter 3.2.
(3) L2 estimates yield u ∈ H 2,2 (Ω) and by De Giorgi-Nash we obtain u ∈
C 1,λ , 0 < λ < 1.
(4) Schauder theory then yields C 2,α estimates.

73
(v) We now employ the Leray-Schauder fix point theorems, cf. next chapter,
to obtain a solution. Let 0 < σ < 1 and consider
∂ãi
Ãu + σã(·, u, Du) + (1 − σ) (·, u, Du) + γ(u − σu0 ) = 0
∂xi (2.15)
u|∂Ω = σφ.

For this equation we also have to prove C 2,α bounds. Choosing γ large
enough, then (2.15) is also coercitive and we obtain estimates independently
of σ. Leray-Schauder then implies , that there is a solution for σ = 1.

2.4.6 Proposition. Let ai , a ∈ C 1 (Ω̄ × R × Rn ) and let u0 ∈ C 0,1 (Ω̄) be a

weak solution of

Au + a(·, u, Du) = 0
u|∂Ω = φ ∈ H 2,p (Ω), p > n.

Then we have
u0 ∈ H 2,p (Ω).

Proof. The same proof as the one of the preceding theorem is applicable.
However, we have to use the Lp -theory of Calderon-Zygmund instead of
Schauder theory.

74
 Chapter 3

Quasilinear operators and

Leray-Schauder theory

3.1 Fixed point theorems, Leray-Schauder theorem

and applications
3.1.1 Theorem. (Schauder’s fixed point theorem)
Let V be a Banach space, K ⊂ V compact and convex and T : K → K
continuous. Then T has a fixed point.

Proof. We use Brouwer’s fixed point theorem. Let k ∈ N, then there exist
(ui )1≤i≤N , ui ∈ K, such that
N
[
K⊂ B 1 (ui ).
k
i=1

Set
Bi := B 1 (ui ).
k

Let
Sk := conv(u1 , ..., uN )
and define

Jk (u) : K → Sk
PN
dist(u, K\Bi )ui
u 7→ Pi=1
N
.
i=1 dist(u, K\Bi )

There holds
PN
i=1 dist(u, K\Bi )(ui − u) 1
kJk u − uk = PN <
i=1 dist(u, K\Bi )
k

75
and since Jk ◦ T : Sk → Sk is continuous, it has a fixed point vk . By com-
pactness there is a subsequence vk → v ∈ K. There holds
1
kvk − T vk k = kJk T vk − T vk k < .
k
⇒ v = T v.

3.1.2 Corollary. Let V be a Banach space, K ⊂ V closed and convex and

let T : K → K be continuous and T (K) precompact. Then T has a fixed
point.

Proof. (i) Let A be a precompact set, then conv(A) is also precompact,

because:
Let  > 0, then
N
[
∃xi ∈ A, 1 ≤ i ≤ N : A ⊂ B (xi ).
i=1

Now let y ∈ conv(A), X

y= λ k yk .
k

Then there exist xik : yk ∈ B (xik ) and thus

X X
ky − λk xik k ≤ λk kyk − xik k < .
k k

⇒ ∀y ∈ conv(A) ∃x̄ ∈ conv(xi ) : ky − x̄k < .

N
[
⇒y∈ B2 (xi ),
i=1

since conv(xi ) is precompact.

(ii) Let
¯ (K)) ⊂ K.
C = conv(T
Then T : C → C has a fixed point.

3.1.3 Theorem. (Schaefer)

Let V be a Banach space, T : V → V continuous and compact. Suppose there
is an M > 0, such that for all solutions of

u = σT u, 0 < σ < 1,

so-called quasi fixed points, we have kuk < M, then T has a fixed point.

76
Proof. Without loss of generality we may assume M = 1, for otherwise
consider M −1 T M. Define
(
T u, kT uk < 1
T ∗u = Tu
kT uk , kT uk ≥ 1.

Then
T ∗ : B̄1 → B̄1
is continuous and T ∗ (B̄1 ) is precompact. Thus T ∗ has a fixed point

u = T ∗ (u).

If kT uk > 1 we obtain
1
u= T u,
kT uk
which contradicts the a priori estimate. Thus kT uk ≤ 1 and so

u = T u.

3.1.4 Lemma. Let V be a Banach space and B = B1 (0). Let T : B̄ → V be

continuous, T (B̄) be precompact and T (∂B) ⊂ B̄. Then T has a fixed point
in B̄. If T (∂B) ⊂ B, then the fixed point lies in B.
Proof. Define (
T u, kT uk ≤ 1
T ∗u = Tu
kT uk , kT uk > 1.
Then T ∗ : B̄1 → B̄1 is continuous and T ∗ (B̄1 ) precompact. Thus

∃u ∈ B̄1 : T ∗ u = u.

⇒ T u = u,
for otherwise we had kT uk > 1.

3.1.5 Theorem. (Leray-Schauder)

Let V be a Banach space and T : V × [0, 1] → V continuous and compact.
Suppose
∀u ∈ V : T (u, 0) = 0
and suppose

∃M > 0 ∀0 < σ < 1 : u = T (u, σ) ⇒ kuk < M.

Then
∃u ∈ V : u = T (u, 1).

77
Proof. Without loss of generality let M = 1, i.e.

u = T (u, σ) ⇒ kuk < 1. (3.1)

Let 0 <  ≤ 1 and let T ∗ : B1¯(0) → V be defined by

(  
u 1−kuk
T , , 1 −  ≤ kuk ≤ 1
T ∗ u = T∗ u = kuk 
u
T ( 1− , 1), kuk ≤ 1 − .

Thus T ∗ is continuous, T ∗ (B̄1 ) is precompact and T ∗ (∂B) = {0}.

Thus there exists u such that

u = T ∗ u .

Defining (
−1 (1 − ku k), 1 −  ≤ ku k ≤ 1
σ =
1, ku k < 1 − ,
we obtain (  
T kuu k , σ , 1 −  ≤ ku k ≤ 1
u =
u
T ( 1− , σ ), ku k < 1 − .
 → 0 implies that for a subsequence we have

(u , σe ) → (u, σ), 0 ≤ σ ≤ 1.

There clearly holds

u = T (u, σ).
Furthermore σ = 1, for otherwise we would find

ku k → 1

and thus
kuk = 1,
which is a contradiction.

3.1.6 Theorem. Let Ω b Rn be open with ∂Ω ∈ C 2,α , 0 < α < 1. Let

ai ∈ C 2 (Ω̄ × R × Rn ), a ∈ C 1 (Ω̄ × R × Rn ), ai elliptic and φ ∈ C 2,α (Ω̄).
Suppose that for all 0 < σ < 1 and for all solutions of the boundary value
problem

∂ai
Au + σa(·, u, Du) + (1 − σ) (·, u, Du) = 0
∂xi (3.2)
u|∂Ω = σφ

there holds
|u| + |Du| ≤ M.

78
Then the Dirichlet problem

Au + a(·, u, Du) = 0
(3.3)
u|∂Ω = φ

has a solution u ∈ C 2,α (Ω̄).

Proof. Let v ∈ C 1,α (Ω̄). Consider the equation

∂ai ∂ai
−aij (·, v, Dv)uij − (·, v, Dv)ui − i (·, v, Dv) + a(·, v, Dv) = 0.
∂u ∂x
Write
∂ai
Lu = −aij (·, v, Dv)uij − (·, v, Dv)ui .
∂u
Then L is a uniformly elliptic differential operator with hoelder continuous
coefficients. From Schauder theory we conclude, that the boundary value
problem
∂ai
Lu + a(·, v, Dv) + (·, v, Dv) = 0
∂xi (3.4)
u|∂Ω = φ

has a solution u ∈ C 2,α (Ω̄) and

|u|2,α,Ω ≤ c(|φ|2,α + |v|1,α,Ω ),

where c = c(λ, |a|1,Ω̄×[−|v|0 ,|v|0 ]×[−|Dv|0 ,|Dv|0 ] , |ai |2,Ω̄×[−|v|0 ,|v|0 ]×[−|Dv|0 ,|Dv|0 ] ).
Define

T : C 1,α (Ω̄) → C 2,α (Ω̄)

v 7→ u = T v,

where u is a solution of (3.4).

T is compact: Let (v k ) be bounded, then uk = T v k is bounded. Thus we
obtain subsequences, such that

v k → v in C 1

and
uk → u in C 2 .
⇒ u = Tv
and by uniqueness the whole sequences must converge.
T is continuous: Write (3.4) in the form

Luk = f k
u|∂Ω = φ.

79
Let v k → v and denote the uk to be the corresponding solutions. Then
uk − ul solves
∂ai
aij (·, v k , Dv k )(uk − ul )ij − (·, v k , Dv k )(uk − ul )i
∂u
+(aij (·, v l , Dv l ) − aij (·, v k , Dv k ))ulij
∂ai ∂ai
+( (·, v l Dv l ) − (·, v k , Dv k ))uli
∂u ∂u
≡f k − f l + F kl
and thus
|uk − ul |2,α ≤ c(|f k − f l |0,α + |F kl |0,α ) → 0.
We have to show that all quasi fixed points are a priori bounded. So let
u = σT u, 0 < σ < 1. This means
∂ai ∂ai
−aij (·, u, Du)uij + σa(·, u, Du) + σ (·, u, Du) + (·, u, Du)ui = 0
∂xi ∂u
u|∂Ω = σφ.
By assumption we have |u| + |Du| ≤ M1 . Thus by the L2 estimates and
DeGiorgi-Nash we find
|u|1,λ ≤ M2 .
Schauder implies
|u|2,λ ≤ M3
and repeating those arguments we find
|u|1,α ≤ M4 .
Setting
M = M4 + 1
implies the claim.

3.2 Gradient bounds

3.2.1 Theorem. Let ai , a be the coefficients of the modified operator in the
proof of Theorem 2.4.5.
Au + a + γ(u − u0 ) ≡ −(ai (·, u, Du))i + a(·, u, Du) + γ(u − u0 ),
∂a ∂a i i
ai ∈ C 1 , a ∈ C 0 , ai uniformly elliptic, a = 0 for |Du| > M and ( ∂x i , ∂u ) = 0
2 2
for |Du| > 1. Let ∂Ω ∈ C , φ ∈ C (Ω̄) and for u ∈ H (Ω) 1,2

Au + a + γ(u − u0 ) = 0
u|∂Ω = φ,
then |Du| ≤ c.

80
Proof. The L2 estimates imply u ∈ H 2,2 (Ω).
Let |Du|∂Ω ≤ k0 , then
|Du| ≤ c(k0 , ...).
Let 1 ≤ k ≤ n and v = uk . Differentiate the equation for xk to obtain

∂ai ∂ai
−(aij uj )i + v + i + Dk a + γ(v − v0 ) = 0.
∂u ∂x
Multiply this equation by

vk := max(v − k, 0),

where k > k0 .
ˆ ˆ ˆ
⇒ aij Di vDj vk + γ vvk ≡ f vk , f ∈ L∞ .
Ω Ω Ω

By the Stampacchia method we obtain

v ≤ k0 + d

and analogueously from below.

Bounds up to the boundary: Choose a tubular neighborhood Ω with 0 ≤
d ∈ C 2 (Ω̄ ). Define an upper barrier w ≡ w+ by

w = φ + Λh(d), 0 ≤ d ≤ .

⇒ wij = φij + Λh0 dij + Λh00 di dj .

−aij uij = f ∈ L∞ .
Choose h(d) = log(1 + αd), where α is large. Choose  = 1
α. Then h00 is the
dominant term and thus
−aij wij > f.
w|{d=} = φ + Λlog 2 > u
⇒ u ≤ w.
Bound it from below by using w− = φ − Λh(d).

81