Institute for Analysis

PD Dr. Rainer Mandel

Dr. Zoïs Moitier
July 23, 2021

Sobolev spaces: tutorials

Exercise sheet 07 with solution

Exercise 1.
Let Ω ⊂ RN be a bounded Lipschitz domain. Show that there is no trace operator such that
W1,N (Ω) → L∞ (∂Ω).
(Clue: In the past, the answers shine)

Solution 1.
We take the same function u as in Exercise sheet 04, Exercise 1, Question 4, to recall this
function is u : RN → R defined by

φ : x 7→ |ln (|x|)|δ χ(x),

 
1 ∞
for 0 < δ < 1 − N
and χ ∈ C0 B 1 (0) such that χ ≡ 1 on B 1 (0). This function has the
2 4

property that φ ∈ W1,N (RN ) and φ 6∈ L∞ (RN ) because of the singularity at x = 0. We define
the sequence of smooth function ψn : Ω → R by

ψn (x) = φ(x − xn ), ∀x ∈ Ω,

N
for a sequence (xn )n∈N ∈ RN \ Ω such that xn → x∗ ∈ ∂Ω.
By contradiction, if we have a continuous trace operator γ : W1,N (Ω) → L∞ (∂Ω) that mean
there exists a contant C > 0 such that

kγukL∞ (∂Ω) ≤ C kukW1,N (Ω) , ∀u ∈ W1,N (Ω),

and the trace correspond to the boundary restriction for function in W1,N (Ω) ∩ C(Ω). We
compute
max |ψn | = kγψn kL∞ (∂Ω) ≤ C kψn kW1,N (Ω) ≤ C kφkW1,N (RN ) .
∂ω

The right hand side is bounded but the left hand part is unbounded as n → +∞ which is a
contraction, therefore, there is no trace operator such that W1,N (Ω) → L∞ (∂Ω).

Exercise 2.
Let Ω ⊂ RN be a bounded Lipschitz domain, a ∈ C(Ω) a nontrivial function with a ≥ 0,
b ∈ C(∂Ω) a nontrivial function with b ≥ 0, and 1 ≤ p < ∞. Show that the following quantities
are norms and equivalent to the k·k1,p norm on W1,p (Ω):
Z Z  p1
p p
1. kuka = |∇u(x)| dx + a(x) |u(x)| dx ;
Ω Ω
Z Z  p1
p p
2. kukb = |∇u(x)| dx + b(x) |γu(x)| dσ(x) .
Ω ∂Ω

1
 (Hint: Theorem 10.5)

Solution 2.
The two norms k·ka and k·kb are non-negative, homogeneous of degree 1, and satisfy the trian-
gular inequality via the Minkowski’s inequality.
Question 1. For the definiteness: if kuka = 0 then k∇ukp = 0 which implies that u is constant
and Ω a |u|p dx = 0 implies that the constant is zero.
R

Now we show that the norm k·ka is equivalent to k·k1,p . We directly have
 
1
kuka ≤ max 1, max a p kuk1,p .
Ω

We show the converse inequality kuk1,p ≤ Ca kuka by contradiction. We assume that there
N
exists a sequence (un )n∈N ∈ W1,p (Ω) such that

kun k1,p = 1 and lim kun ka = 0.

n→+∞

The sequence (un )n∈N is bounded in W1,p (Ω) so by the Rellich-Kondrachov theorem, there exists
a converging subsequence uσ(n) → u in Lp (Ω). Furthermore, limn→+∞ uσ(n) a = 0 implies that
limn→+∞ ∇uσ(n) p = 0 and, for φ ∈ C0∞ (Ω) and i ∈ {1, . . . , N }, we compute
Z Z Z
0 = lim ∂i uσ(n) (x)φ(x) dx = − lim uσ(n) (x)∂i φ(x) dx = − u(x)∂i φ(x) dx
n→+∞ Ω n→+∞ Ω Ω

therefore u ≡ c a constant. From limn→+∞ kun ka = 0, we deduce

Z Z
p
0 = lim a(x) uσ(n) dx = a(x) |c|p dx
n→+∞ Ω Ω

which give u ≡ c = 0 and, from kun k1,p = 1, we have the contradiction

p p
1 = ∇uσ(n) p + uσ(n) p .
| {z } | {z }
→0 →0

So by contradiction, there exists a constant Ca > 0 such that kuk1,p ≤ Ca kuka for all u ∈
W1,p (Ω) and we have
  p1
Ca−1 kuk1,p ≤ kuka ≤ max 1, max a kuk1,p .
Ω

Question 2. For the definiteness: if kukb = 0 then k∇ukp = 0 which implies that u is constant
and ∂Ω b |γu|p dσ(x) = 0 implies that the constant is zero. Now we show that the norm k·kb is
R

equivalent to k·k1,p . We directly have

  p1
kukb ≤ 1 + |||γ|||pW1,p (Ω)→Lp (∂Ω) max b kuk1,p .
∂Ω

We show the converse inequality kuk1,p ≤ Cb kukb by contradiction. We assume that there
N
exists a sequence (un )n∈N ∈ W1,p (Ω) such that

kun k1,p = 1 and lim kun kb = 0.

n→+∞

2
The sequence (un )n∈N is bounded in W1,p (Ω) so by the Rellich-Kondrachov theorem, there exists
a converging subsequence uσ(n) → u in Lp (Ω). Furthermore, limn→+∞ uσ(n) b = 0 implies that
limn→+∞ ∇uσ(n) p = 0 and, for φ ∈ C0∞ (Ω) and i ∈ {1, . . . , N }, we compute
Z Z Z
0 = lim ∂i uσ(n) (x)φ(x) dx = − lim uσ(n) (x)∂i φ(x) dx = − u(x)∂i φ(x) dx
n→+∞ Ω n→+∞ Ω Ω

therefore u ≡ c a constant. From limn→+∞ kun kb = 0, we deduce

Z Z
p
0 = lim b(x) γuσ(n) dσ(x) = b(x) |c|p dx
n→+∞ Ω Ω

which give u ≡ c = 0 and, from kun k1,p = 1, we have a contradiction

p p
1 = ∇uσ(n) p + uσ(n) p .
| {z } | {z }
→0 →0

So by contradiction, there exists a constant Cb > 0 such that kuk1,p ≤ Cb kukb for all u ∈
W1,p (Ω) and we have
  p1
Cb−1 kuk1,p ≤ kukb ≤ 1 + |||γ|||pW1,p (Ω)→Lp (∂Ω) max b kuk1,p .
∂Ω

Exercise 3.
Let Ω ⊂ RN be a bounded Lipschitz domain, f ∈ L2 (Ω) a function, and α ∈ C(∂Ω) a nontrivial
function with α(x) ≥ 0 for almost every x ∈ ∂Ω. We denote a : H1 (Ω) × H1 (Ω) → R a bilinear
form, and ` : L2 (Ω) → R a linear form define by
Z Z
a(u, v) := ∇u(x) · ∇v(x) dx + α(x) γu(x) γv(x) dσ(x),
ZΩ ∂Ω

`(v) := f (x) v(x) dx.

Ω

1. Show that there exists u ∈ H1 (Ω) such that a(u, v) = `(v), for all v ∈ H1 (Ω).

2. Assume that the solution u of Question 1 is in H2 (Ω). Find the PDE that u satisfies.

Solution 3.
Question 1. From previous exercise a is coercive. From the trace theorem a and ` are
continues. By the Lax-Milgram theorem, there exists a unique u ∈ H1 (Ω) such that a(u, v) =
`(v), for all v ∈ H1 (Ω).

Question 2. For v ∈ H1 (Ω), we have a(u, v) = `(v) which give

Z Z Z
∇u(x) · ∇v(x) dx + α(x) γu(x) γv(x) dσ(x) = f (x) v(x) dx
Ω ∂Ω Ω

we deduce the expression

Z Z
[−∆u(x) − f (x)] v(x) dx + [∂n u(x) + α(x)γu(x)] γv(x) dσ(x) = 0 (1)
Ω ∂Ω

3
with ∂n the exterior normal derivative. First take v ∈ C0∞ (Ω), Eq. (1) become
Z
[−∆u(x) − f (x)] v(x) dx = 0
Ω

and −∆u(x) − f (x) ∈ L2 (Ω), therefore, we obtain −∆u(x) = f (x), for almost everywhere
x ∈ Ω. Now, taking w ∈ C ∞ (∂Ω) and v ∈ H1 (Ω) such that w = γv, Eq. (1) reduce to
Z
[∂n u(x) + γu(x)] w(x) dσ(x) = 0
∂Ω

and ∂n u(x) + γu(x) ∈ L2 (∂Ω) therefore ∂n u(x) + γu(x) = 0, for almost everywhere x ∈ ∂Ω. So
the PDE is (
−∆u = f in Ω
.
∂n u + αu = 0 in ∂Ω
Remark 1. A boundary condition of the form β∂n u + αu = g is usually referred as a Robin
boundary condition or impedance boundary conditions.