2019/2020

Master 2 Recherche fondamentale en mathématiques

Sobolev spaces’s exercises

Professor : Roger Lewandowski

Written by Pierre Le Barbenchon

1 Sobolev spaces
1.1 General framework

Exercise 1. N = 2

1) Define the outward unit vector at any point x

Z
2) Define g(x)dσ(x) for g smooth
Γi

3) Prove (IP ) by using 1D integration by part and Fubini’s theorem.

y Γ = Γ1 ∪ Γ2
→
n
z+ Γ1
> y = f1 (x)

< y = f2 (x)

Γ2
x
a b

Solution. 1) The tangent of Γ1 on x has (1, f10 (x)) as director vector. Then, the normal vector is
(−f10 (x), 1) because the inner product has to be zero. We want to have a outward unit vector,
so we divide by the norm and find

−f 0 (x) 1
~n(1) = ( p 0 1 ,p 0 )
2
(f1 (x)) + 1 (f1 (x))2 + 1

Same method for Γ2 and we find :

−f 0 (x) 1
~n(2) = ( p 0 2 ,p 0 )
2
(f2 (x)) + 1 (f2 (x))2 + 1

1
2) We parameterize Γ1 with (x, f1 (x)) for x ∈ [a, b]. Then we use line integral.
Z Z b
g(x)dσ(x) = g(f1 (x))||f10 (x)||dx
Γ1 a
s 
dx 2 df1 (x) 2
Z b  
= g(f1 (x)) + dx
a dx dx
Z b q
= g(f1 (x)) 1 + (f10 (x))2 dx
a

Same method for Γ2 :

Z Z a q
g(x)dσ(x) = g(f2 (x)) 1 + (f20 (x))2 dx
Γ2 b

We change the bounds order because we go from b to a on the boundary Γ2 .

3) We use the Fubini theorem to split the open Ω

!
Z Z b Z f1 (x)
∂v ∂v
u(x, y) (x, y)dxdy = u(x, y) (x, y)dy dx
Ω ∂y x=a y=f2 (x) ∂y

Then we use integration by parts :

!
Z b Z f1 (x)
f (x) ∂u
= [u(x, y)v(x, y)]f12 (x) − (x, y)v(x, y)dy dx
x=a f2 (x) ∂y
!
Z b Z f1 (x)
∂u
= u(x, f1 (x))v(x, f1 (x)) − u(x, f2 (x))v(x, f2 (x)) − (x, y)v(x, y)dy dx
x=a f2 (x) ∂y

Z b
1
q
= u(x, f1 (x))v(x, f1 (x)) p 1 + (f10 (x))2 dx
x=a 1 + (f10 (x))2
Z b
1
q
− u(x, f2 (x))v(x, f2 (x)) p 1 + (f20 (x))2 dx
x=a 1 + (f20 (x))2
Z b Z f1 (x)
∂u
− (x, y)v(x, y)dydx
x=a f2 (x) ∂y
Z b q
(1)
= uvn2 1 + (f10 (x))2 dx
x=a
Z a q
(2)
+ uvn2 1 + (f20 (x))2 dx
x=b
Z b Z f1 (x)
∂u
− (x, y)v(x, y)dydx
x=a f2 (x) ∂y
Z Z Z
(1) (2) ∂u
= uvn2 dσ + uvn2 dσ − (x, y)v(x, y)dydx
Γ1 Γ2 Ω ∂y
Z Z
∂u
= uvn2 dσ − (x, y)v(x, y)dydx
Γ Ω ∂y

2
 We finally find the (IP ) formula. (We justify the Fubini Theorem because u, v ∈ C 1 , then
∂v
they are continuous on Ω compact, then u ∈ L1 (Ω))
∂y

Exercise 2. Let Ω =] − 1; 1[⊂ R.

x
 if x 6= 0
sgn(x) = |x| ∈ L∞ (Ω)
0 if x = 0

Show that x 7→ sgn(x) has no weak derivative.

Solution. We work in the distributions. For all ϕ ∈ D(] − 1, 1[),

hsgn0 , ϕi = −hsgn, ϕ0 i
Z 0 Z 1
0
=+ ϕ − ϕ0
−1 0
= [ϕ]0−1 − [ϕ]10
= 2ϕ(0)
= h2δ0 , ϕi

Then sgn0 = 2δ0 . But this distribution isn’t in L1loc , so x 7→ sgn(x) has no weak derivative.

To prove that the dirac δ0 is not in L1loc , we work for a contradiction. Let f ∈ L1loc such that
Z 1
∀ϕ ∈ D(] − 1, 1[), ϕ(0) = fϕ
−1

We use a sequence ϕn ∈ D(] − 1, 1[) such that supp(ϕn ) ⊂ [− n1 , n1 ], ϕn (0) = 1 and the supremum
of each ϕn equal to 1. R1
Then we have by the dominated convergence theorem that −1 f ϕn goes to 0 because we can
dominate |1[− 1 , 1 ] f ϕn | by |f |1[−1,1] which is in L1 .
n n
We have a contradiction because ϕn (0) = 1 for all n ∈ N.

Exercise 3. Let U1 and U2 be two opensets in Rn such that U1 ∩ U2 6= ∅. Let Ω = U1 ∪ U2 ,

α = (α1 , . . . , αn ) a multiindex, and u ∈ L1loc (Ω)
Assume that u has a weak derivative v1 = Dα u in U1 and v2 = Dα u in U2 .

1) Prove that v1 = v2 in U1 ∩ U2
(
v1 in U1
2) Let v = .
v2 in U2
Prove that Dα u exists in Ω and Dα u = v.

3
Solution. 1) Soit ϕ ∈ D(U1 ∩ U2 ),
Z Z
hu, Dα ϕi = uDα ϕ = − v1 ϕ
U1 ∩U2 U1 ∩U2

and Z Z
α α
hu, D ϕi = uD ϕ = − v2 ϕ
U1 ∩U2 U1 ∩U2
R
Then U1 ∩U2 (v2 − v1 )ϕ = 0 for all ϕ ∈ D(U1 ∩ U2 ). Hence, we have v1 = v2 in U1 ∩ U2 . We
just use the injectivity of (
L1loc (Ω) → D0 (Ω)
f 7→ Tf

2) Let ϕ ∈ D(Ω). We take a unit partition of Ω = U1 ∪ U2 relatively to supp(ϕ). Then

we have θ1 , θ2 ∈ D(Ω) with θ1 + θ2 = 1, 0 6 θ1 , θ2 6 1, supp(θ1 ) ⊂ U1 ∩ supp(ϕ) and
supp(θ2 ) ⊂ U2 ∩ supp(ϕ)
We have ϕ = θ1 ϕ + θ2 ϕ

hDα u, ϕi = hDα u, θ1 ϕi + hDα u, θ2 ϕi

Z Z
= v 1 θ1 ϕ + v2 θ2 ϕ because suppθ1,2 ⊂ U1,2
U1 U2
Z Z
= vθ1 ϕ + vθ2 ϕ
ZU1 U2
Z
= vθ1 ϕ + vθ2 ϕ by definition of θ1 and θ2 supports
U1 ∪U2 U1 ∪U2
Z
= v(θ1 + θ2 )ϕ
U1 ∪U2
= hv, ϕi

Hence Dα u exists in Ω and we have Dα u = v in Ω.

1.2 Definition and basic properties of W m,p (Ω)

Exercise 4. Let Ω = B(0, 1) ⊂ R2 .


Ω →R
u:  1 
x 7→ ln ln
|x|

Prove that u ∈ H 1 (Ω) (but u ∈

/ C 0 (Ω))

4
 Remark :
I just show the result for H 1 (B(0, 21 )) because I didn’t succeed for u0 ∈ L2 (B(0, 1)) (since it’s
not true), but it keeps the spirit of the exercise because we have a function in H 1 (B(0, 21 )) which
is not in C 0 (B(0, 12 )).

u(x, y)

−1
x
0
1
−1

/ C 0 (Ω). And we can’t find a continuous

Solution. The function u is not continuous in (0, 0), so u ∈
representative of u (somebody continuous equal to u almost everywhere).

∂ ∂
We want to prove that u ∈ L2 (Ω) and u, u ∈ L2 (Ω).
∂x ∂y

Step 1 : Let start with u ∈ L2 (Ω).

Z Z 1 Z 2π
2 1 2 
|u(x)| dx = ln
ρdθdρ ln
Ω ρ=0 θ=0 ρ
Z 1 
1 2
= 2π ln ln ρdρ
0 ρ

while doing a change of coordinates and seeing that ln ρ1 is non negative for ρ ∈]0, 1[.
We use the inequality ln(x) 6 x − 1 for x > 0.
Z Z 1 1 2
2
|u(x)| dx = 2π ln − 1 ρdρ
Ω 0 ρ
Z 1
= 2π (− ln(ρ) − 1)2 ρdρ
0
Z 1
= 2π (ln(ρ) + 1)2 ρdρ
0

5
 We use an other change of coordinates : ρ = exp(−x)
Z Z +∞
2
|u(x)| dx = 2π (1 − x)2 e−2x dx < +∞
Ω 0
We know that x 7→ (1 − x)2 e−2x is continuous on R+ and summable in +∞ and 0. Hence,
u ∈ L2 (Ω)
Step 2 : We want to show that ∇u !! ∈ L2 (B(0, 21 )) (B(0, 21 ) as I said in the previous remark).
1
Let u(x, y) = ln ln p for all (x, y) ∈ B(0, 12 ).
x + y2
2

Assume for the moment that the weak derivative is equal to the classical derivative.
We have
∂ −2x 1 x
u(x, y) = 1
p .  = p
∂x √ 2( x2 + y 2 )3/2 ln √ 1 (x2 + y 2 ) ln( x2 + y 2 )
2
x +y 2
2
x +y 2

and
∂ y
u(x, y) = p
∂y (x + y ) ln( x2 + y 2 )
2 2

Let compute the L2 norm.

Z Z
∂ 2 ∂
u(x, y) dxdy + u(x, y)2 dxdy
B(0,1/2) ∂x B(0,1/2) ∂y
x2 y2
Z
= 2
p + p dxdy
B(0,1/2) (x2 + y 2 )2 ln ( x2 + y 2 ) (x2 + y 2 )2 ln2 ( x2 + y 2 )
Z
1
= 2
p dxdy
B(0,1/2) (x + y ) ln ( x2 + y 2 )
2 2
Z 2π Z 1/2
1
= 2 2
ρdρdθ
θ=0 ρ=0 ρ ln ρ
Z 1/2
1
=2π 2 dρ we use the change of variables ρ = ex
ρ=0 ρ ln ρ
Z − ln(2) x
e
=2π x 2
dx because dρ = ex dx
x=−∞ e x
Z − ln(2)
1
=2π 2
dx < +∞
x=−∞ x

We see in the last line why I work in B(0, 12 ).

It proves that ∇u ∈ L2 (B(0, 21 )) and then u ∈ H 1 (B(0, 12 )).
Step 3 : We have to justify that the weak derivative is equal to the classical derivative.
Let take ϕ ∈ D(B(0, 1)),
∂ ∂
h u, ϕi = −hu, ϕi
∂x Z ∂x
∂
=− u ϕ
B(0,1) ∂x
Z
∂
= −lim u ϕ
→0 B(0,1)\B(0,) ∂x

6
 We can apply the Green Formula
Z Z Z Z
∂ ∂
u ϕ=− uϕ + uϕnx dσ + uϕnx dσ
B(0,1)\B(0,) ∂x B(0,1)\B(0,) ∂x S(0,) S(0,1)

But ϕ ∈ D(B(0, 1)), so ϕ vanishes on S(0, 1)1 .

R
And S(0,) uϕnx dσ 6 kuk∞,S(0,) kϕk∞,S(0,) 2π because 2π is the perimeter of a cercle with
radius 1. This quantity goes to zero when  goes to zero.
It follows that
Z
∂ ∂
h u, ϕi = −lim − uϕ
∂x →0 B(0,1)\B(0,) ∂x
Z
∂
= lim uϕ
→0 B(0,1)\B(0,) ∂x
Z
x
= lim p ϕ(x, y)dxdy
→0 B(0,1)\B(0,) (x + y ) ln( x2 + y 2 )
2 2
Z
x
= p ϕ(x, y)dxdy
B(0,1) (x + y ) ln( x2 + y 2 )
2 2

while using the dominated convergence theorem.

∂
We can do the same method for u, then the weak derivative is equal to the classical one.
∂y

Exercise 5. Let (ri )i be a countable and dense set in B(0, 1) ⊂ RN , α > 0.

+∞
X 1
u(x) = |x − ri |−α
2i
i=0

For p > 1, prove that u ∈ W 1,p (B(0, 1)) for all α < α0 with α0 = α0 (N, p) to be calculated.
1
(Hint : start by studying x 7→ )
|x|α

Remark :
I will prove the result for α0 = Np − 1 and that is a sufficient condition as the exercise asks.
But it’s a necessary and sufficient condition.

Solution. Z
1
Step 1 : We want to show that α
dx < ∞ if and only if α < N . To compute
Z B(0,1) |x|
1
α
dx, we use a spherical coordinates change in dimension N .
B(0,1) |x|
1
S(0, 1) represents the sphere of radius 1

7
 x1 = ρ cos(θ1 )
x2 = ρ sin(θ1 ) cos(θ2 )
x3 = ρ sin(θ1 ) sin(θ2 ) cos(θ3 )
.. ..
. .
xN −1 = ρ sin(θ1 ) . . . sin(θN −2 ) cos(θN −1 )
xN = ρ sin(θ1 ) . . . sin(θN −2 ) sin(θN −1 )

The spherical volume element is

dV = ρN −1 sin(θ1 )N −2 sin(θ2 )N −3 . . . sin(θN −3 )2 sin(θN −2 )dρdθ1 . . . dθN −1

Z
1
dx
B(0,1) |x|α
Z1 Z π Z 2π
1
= dV
ρ=0 θ1,...,N −2 =0 θN −1 =0 ρα

But the following quantity is finite, let call it K.

Z π Z 2π
K= sin(θ1 )N −2 sin(θ2 )N −3 . . . sin(θN −3 )2 sin(θN −2 )dθ1 . . . dθN −1
θ1,...,N −2 =0 θN −1 =0

Then the computation gives

Z
1
dx
B(0,1) |x|α
Z 1
1 N −1
=K ρ dρ
ρ=0 ρα
Z 1
1
=K dρ
ρ=0 ρα−N +1

The integral converges if and only if α − N + 1 < 1. Then

Z
1
α
dx < ∞ if and only if α < N
B(0,1) |x|

1
Step 2 : Let introduce fα : RN → R such that f (x) = . Thank to the step 1, we have
|x|α
fα ∈ Lp (B(0, 1)) if and only if pα < N .
N
So we have fα ∈ Lp (B(0, 1)) if and only if α < .
p

8
Step 3 : We want to show that u ∈ Lp (B(0, 1)) if α < Np .
We start by using the Fatou’s Lemma with the functions
n
!p
X 1
hn (x) = |x − ri |−α
2i
i=0

We have Z Z
lim inf hn 6 lim inf hn

While passing to the power p1 , we have

Z 1 Z 1
p p
lim inf hn 6 lim inf hn

n
!p !1 n
!p !1
Z p Z p
X 1 X 1
lim inf |x − ri |−α dx 6 lim inf |x − ri |−α dx
2i 2i
i=0 i=0
Z +∞
!p ! p1 n
X 1 X 1
|x − ri |−α dx 6 lim inf fα (. − ri )
2i 2i
i=0 i=0 p
n
X 1
kukp 6 lim inf kfα (. − ri ) kp
2i
i=0
+∞
X 1
kukp 6 kfα (. − ri ) kp
2i
i=0

We used the Minkowski inequality to get out the sum of the norm Lp .
But thanks to step 2, while using a translation of ri , we have
N
kfα (. − ri )kp < +∞ if and only if α <
p
And all those quantities are dominated by the same function g = fα on B(0, 2), so
+∞
X 1
kfα (. − ri ) kp < +∞
2i
i=0

N
Then u ∈ Lp if α < .
p
Step 4 : We want to show that u0 ∈ Lp (B(0, 1)).
We start to find the derivative of fα (. − r) for r = (r1 , . . . , rN ) ∈ B(0, 1). For the same reasons
of the Exercise 4 Step 3, the weak derivative is equal to the classical derivative.
∂ ∂ 1
fα (. − r) =
∂xk ∂xk ( (xj − rj )2 )α/2
P

α −1
− α2 2(xk − rk ) (xj − rj )2 2
P
=
|x − r|2α
−α(xk − rk )
=
|x − r|α+2

9
 ∂ α N
Then we have fα (. − r) < . And it’s in Lp if α + 1 < p. (because of the Step 2)
∂xk |x − r|α+1

We want to show that

+∞
! +∞
∂ X 1 X 1 ∂
i
fα (. − ri ) = fα (. − ri ) (1)
∂xk 2 2i ∂xk
i=0 i=0

For all ϕ ∈ D(B(0, 1)),

+∞
* ! + * +∞ +
∂ X 1 X 1 ∂
fα (. − ri ) , ϕ = fα (. − ri ), ϕ
∂xk 2i 2i ∂xk
i=0 i=0
Z X+∞
1 ∂
= fα (. − ri ) ϕ
2i ∂xk
i=0
+∞ Z
X 1 ∂
= i
fα (. − ri ) ϕ
2 ∂xk
i=0
+∞  
X 1 ∂
= fα (. − ri ), ϕ
2i ∂xk
i=0
+∞  
X 1 ∂
= fα (. − ri ), ϕ
2i ∂xk
i=0

Z X
The justification of the change of and is because of the dominated convergence theorem.

n +∞
X 1 ∂ X 1 ∂
i
fα (. − ri ) ϕ −−−→ i
fα (. − ri ) ϕ a.e.
2 ∂xk n→∞ 2 ∂xk
i=0 i=0

and
n ∞
X 1 ∂ ∂ X 1
fα (. − ri ) ϕ 6 ϕ 1Supp ϕ fα (. − ri )
2i ∂xk ∂xk ∞ 2i
i=0
|i=0 {z }
∈Lp (Step 3)
| {z }
∈L1 (Supp ϕ)

Then it justifies the egality (1) and so, thanks the same methods that Step 3 and the beginning
of Step 4, we have
+∞
!
∂ X 1 N
fα (. − ri ) < +∞ if and only if α < −1
∂xk 2i p
i=0 p

Step 5 : Conclusion
N
Thanks to Step 3 and 4, we have u ∈ W 1,p (B(0, 1)) for all α < α0 with α0 = − 1.
p

10
1.3 Duality spaces W −m,p

Exercise 6. Let α = (α1 , . . . , αN ), x = (x1 , . . . , xN ). Case of Ω = RN .

S := {u ∈ L2 (RN ) such that xα Dβ u ∈ L2 (RN )∀α, β multiindex}

Show that
H m (RN ) = {u ∈ S 0 , such that (1 + |ξ|2 )m/2 û ∈ L2 (RN )}
where û is the Fourier transform of u.

Solution. We recall that

H m (RN ) := {u ∈ L2 (RN ), Dα u ∈ L2 (RN ), ∀|α| 6 m}

Step 1 : We wantPto remark that F(Dα u)(ξ) = i|α| ξ α F(u)(ξ) for u ∈ S, α = (α1 , . . . , αN ) ∈ NN ,
αN
ξ ∈ RN with |α| = i αi and ξ α = ξ1α1 . . . ξN .
For ϕ ∈ S,

hF(Dα u), ϕi = hDα u, ϕi

= (−1)|α| hu, Dα (Fϕ)i
= (−1)|α| hu, (−i)|α| ξ α Fϕi
= i|α| ξ α hu, Fϕi
= i|α| ξ α hFu, ϕi

Then we have F(Dα u)(ξ) = i|α| ξ α F(u)(ξ).

Step 2 : We want to prove the direct inclusion.

Let u ∈ H m (RN ), then we have by Fourier–Plancherel, F(Dα u) ∈ L2 for all α such that |α| 6 m.
But Dα u ∈ S 0 , so i|α| ξ α û ∈ L2 (by Step 1).

Z  2 Z
m
2
(1 + |ξ| ) û(ξ)
2 dξ = (1 + |ξ|2 )m û2 (ξ)dξ
RN RN
Z m  
X m
= |ξ|2k û2 (ξ)dξ
RN k=0 k
m  Z
X m
= |ξ|2k û2 (ξ)dξ because k 6 m
k N
k=0 |R {z }
<∞

Since u is in S 0 (because L2 ⊂ S 0 ), then

u ∈ {u ∈ S 0 , such that (1 + |ξ|2 )m/2 û ∈ L2 (RN )}

11
Step 3 : We want to prove the other inclusion.
Let u ∈ S 0 such that (1 + |ξ|2 )m/2 û ∈ L2 (RN ).
Let α ∈ NN such that |α| 6 m.
We know that
(1 + |ξ|2 )m û2 (ξ) ∈ L1 (RN ),
then
(1 + |ξ|2 )α û2 (ξ) ∈ L1 (RN ).
It follows that
|ξ|2α û2 (ξ) ∈ L1 (RN ),
then
i|α| ξ α û(ξ) ∈ L2 (RN ).
We recognize the Fourier Transform of Dα u, and, by Fourier–Plancherel, we conclude that
Dα u ∈ L2 (RN ).
Therefore u ∈ H m (RN ).

1.4 Study of W 1,p (Ω)

2
( Ω = {(x, y), 0 < |x| < 1, 0 < y < 1} ⊂ R .
Exercise 7. Let
1 if x > 0
Let u(x, y) =
0 if x < 0

1) Show that u ∈ W 1,p (Ω), ∀p > 1

2) Show there is  > 0, such that there is no function φ ∈ C 1 (Ω) such that ku − φk1,p < .

3) What’s up ?

u(x, y)

x
−1 0 1

12
Solution. 1) Step 1 : We want to show that u ∈ Lp (Ω).
Z 0 Z 1 Z 1 Z 1 Z 1 Z 1
p p
|u(x, y)| dydx + |u(x, y)| dydx = 1dydx = 1
x=−1 y=0 | {z } x=0 y=0 | {z } x=0 y=0
=0 =1

Then u ∈ Lp (Ω).

Step 2 : We want to show that Du ∈ Lp (Ω).

Let ϕ ∈ D(Ω), then since the support of ϕ is compact in Ω, we have ϕ(0, y) = ϕ(1, y) = 0 for
all y ∈ [0, 1] and ϕ(x, 1) = ϕ(x, 0) = 0 for all x ∈ [0, 1].

h∂y u, ϕi = −hu, ∂y ϕi
Z
= − u∂y ϕ
Ω
Z 1 Z 1
=− ∂y ϕ(x, y)dydx
x=0 y=0
Z1
=− (ϕ(x, 1) − ϕ(x, 0))dx
x=0
=0

h∂x u, ϕi = −hu, ∂x ϕi
Z
= − u∂x ϕ
Ω
Z 1 Z 1
=− ∂x ϕ(x, y)dxdy
y=0 x=0
Z 1
=− (ϕ(1, y) − ϕ(0, y))dy
y=0
=0

Then ∂x u(x, y) = ∂y u(x, y) = 0 for all (x, y) ∈ Ω.

So ∂x u, ∂y u ∈ Lp (Ω) and therefore u ∈ W 1,p (Ω).

2) We want to show that there exists ε such that for all φ ∈ C 1 (Ω),

ku − φk1,p,Ω > ε.

We work for a contradiction. We suppose that

1
∀n > 0, ∃φn ∈ C 1 (Ω) s.t. ku − φn k1,p,Ω 6
n

13
So we have a sequence (φn )n such that

W 1,p (Ω)
φn −−−−−→ u
n→∞

It means that
Lp (Ω) Lp (Ω)
φn −−−−→ u and ∇φn −−−−→ 0 (2)
n→∞ n→∞

Because ∂x u(x, y) = ∂y u(x, y) = 0 for all (x, y) ∈ Ω.

By the Lebesgue Inverse Theorem, we have a subsequence, still called φn , that converges to u
almost everywhere in Ω.
We have for almost everywhere y ∈]0, 1[, x1 < 0, x2 > 0

φn (x1 , y) −−−→ u(x1 , y) = 0
n→∞
φn (x2 , y) −−−→ u(x2 , y) = 1
n→∞

Then

|φn (x1 , y) − φn (x2 , y)| −−−→ |u(x1 , y) − u(x2 , y)| = 1 (3)

n→∞

On the other hand,

Z 1
|φn (x1 , y) − φn (x2 , y)| 6 |Dφn (tx1 + (1 − t)x2 , y).(x1 − x2 , 0)|dt
0
Z 1
6 kDφn (tx1 + (1 − t)x2 , y)kk(x1 − x2 , 0)kdt by Cauchy–Schwarz
0
Z 1 1/p
6 k(x1 − x2 , 0)k kDφn (tx1 + (1 − t)x2 , y)kp dt by Hölder
0

We use a convexity inequality

p/2
a2 + b2 xp + y p

6
2 2

In other words, p
(a2 + b2 )p/2 6 2 2 −1 (ap + bp )

Z 1
kDφn (tx2 + (1 − t)x2 , y)kp dt (4)
0
Z 1
= (∂x φ2n (tx1 + (1 − t)x2 , y) + ∂y φ2n (tx1 + (1 − t)x2 , y))p/2 dt (5)
0
Z 1
6C ∂x φpn (tx1 + (1 − t)x2 , y) + ∂y φpn (tx1 + (1 − t)x2 , y)dt (6)
0

14
 We use Fatou to have
Z 1 Z 1
lim inf |φn (x1 , y) − φn (x2 , y)|dy 6 lim inf |φn (x1 , y) − φn (x2 , y)|dy (7)
0 0
| {z }
=1 by (3)

By (6),
Z 1 Z 1Z 1
|φn (x1 , y) − φn (x2 , y)|dy 6 C ∂x φpn (tx1 + (1 − t)x2 , y) + ∂y φpn (tx1 + (1 − t)x2 , y)dtdy
0 Z0 0
6C k∇φn kp −−−−−→ 0 because of (2)
A n→+∞

where A = {(x, y), x1 < x < x2 , 0 < y < 1}

Therefore, by (7), we have a contradiction.
To conclude, we have that there exists ε such that for all φ ∈ C 1 (Ω),
ku − φk1,p,Ω > ε.

3) We have a theorem that says : If Ω is of class C 1 and u ∈ W 1,p (Ω), then there exists a sequence
(un )n of functions in D(RN ) such that
W 1,p (Ω)
un |Ω −−−−−→ u
n→∞

The problem is that, in our exercise, Ω is not of class C 1 and then we can’t use that theorem.

1.4.1 Density results

Exercise 8. 1) Show that ∀v ∈ H m (RN ), ∀m > 1,

kv ? ρ − vkm−1,2 6 Ckvkm,2

2) Show that ∀v ∈ H m (RN ), ∀k > 0,

Cm,k
kv ? ρ km+k,2 6 kvkm,2
k

3) Show that ∀v ∈ H m (RN ), ∀k > 0, ∀|α| 6 k

Ck
kρ ? Dα vk0,∞ 6 N/2+k
kvk0,2


Remark :
In this exercise, for the purposes of notation, I will write sometimes kf (x)kL2 instead of kf kL2
or kf (·)kL2 . It is just because I want to mention the variables sometimes.
Moreover, there is a constant while using Fourier Plancherel due to my definition of the Fourier
transform, but I will never put it (as it was 1). It doesn’t matter because it will be in the constant
C of the exercise.

15
Solution. 1) ∀|α| 6 m − 1, ∀v ∈ H m (RN )

kDα (v ? ρ − v)kL2 = kF(Dα (v ? ρ − v))kL2 with Fourier Plancherel

= kξ α F(v ? ρ − v)kL2 by properties of F
= kξ α F(v)(F(ρ ) − 1)kL2 because of the convolution product

Z
F(ρ )(ξ) = ρ (x)e−iξx dx
RNZ
1 x
= ρ e−iξx dx
N RN 
Z
1
= N ρ(u)e−iξu N du change of variable x = u
 R N

= F(ρ)(ξ)

So we have

F(ρ )(ξ) = F(ρ)(ξ) (8)

R
We remark that 1 = F(ρ)(0) because RN ρ(x)dx = 1.
Now we use the mean value theorem in RN , we have

∃θ ∈]0, 1[, s.t. F(ρ ) − 1 = F(ρ)(ξ) − F(ρ)(0) = DF(ρ)(ξθ).(ξ)

But DF(ρ)(ξθ).(ξ) is bounded by C||ξ where C is a constant.2

To conclude we have

kDα (v ? ρ − v)kL2 6 Ckξ α ξF(v)kL2

6 Ckξ α+1 F(v)kL2
6 CkF(Dα+1 v)kL2
6 CkDα+1 vkL2
6 Ckvkm,2 because |α| + 1 6 m

So we have the result. We just used the notation α + 1, we have to understand that for
instance we write the mean value theorem for the first derivative, then the “+1” is on the first
component of α.
2
Indeed, DF(ρ) = F(xρ) is bounded, because it’s the Fourier transform of a C0∞ function, so we can bound it by
the L1 −norm.

16
2) ∀|α| 6 m, ∀v ∈ H m (RN ), ∀|β| 6 k

kDα+β (v ? ρ )kL2 = kF(Dα+β (v ? ρ ))kL2 with Fourier Plancherel

= kξ α+β F(v ? ρ )kL2 by properties of F
= kξ α+β F(v)F(ρ )kL2 by properties of F and convolution product
(ξ)β
= kξ α F(v) F(ρ)(ξ)kL2 because (8)
|β|
1
= kF(Dα v)F(Dβ ρ)(ξ)kL2 because |β| 6 k
k
C
6 k kF(Dα v)kL2

C
6 k kDα vkL2

C
6 k kvkm,2 because |α| 6 m

So we have the result. We just used C the bound of F(Dβ ρ) that exists since ρ ∈ C ∞ .

3) First we will use the following inequality

∀f, g ∈ L2 , |f ? g(x)| 6 kf kL2 kgkL2

To prove that inequality, we use the definition of the convolution product and the Cauchy–
Schwarz inequality.
So we have, for |α| 6 k,

|ρ ? Dα v(x)| = |Dα (ρ ) ? v(x)| (9)

α
6 kD ρ kL2 kvk0,2 (10)

Ck
We want to prove that kDα ρ kL2 6 N/2+k
.

Z
kDα (ρ )k2L2 = Dα (ρ )(x)2 dx
ZRN

= α (ρ )(ξ)2 dξ by Fourier Plancherel

D\ 
N
ZR
= ξ 2α ρb (ξ)2 dξ
N
ZR
= ξ 2α ρb(ξ)2 dξ because (8)
RN
u2α
Z
1
= 2|α|
ρb(u)2 N du because of the change of variable ξ = u
RN  
1
= kuα F(ρ)(u)k2L2
2k N

17
 So we have
1
kDα (ρ )kL2 = kF(Dα ρ)kL2
k N/2
1
6 N/2+k kDα ρkL2

1
6 N/2+k sup kDα ρkL2
 |α|6k
| {z }
Ck

Then, since (10) and while passing to the supremum, we have the result
Ck
kρ ? Dα vkL∞ 6 N/2+k
kvkL2 .


1.4.2 About traces

Exercise
P 9. Let ~u = (u1 , . . . , uN ) ∈ D(Ω)N with Ω bounded of class C 1 . We recall that div ~u =
i Di ui . We define
p
Hdiv (Ω) = {~u ∈ Lp (Ω)N , div ~u ∈ Lp (Ω)}
p
1) Show that Hdiv (Ω) is a Banach space.

2) Prove that ~u.~n|Γ can be defined in an appropriate space to be determined. (~u.~n|Γ = “normal
trace of u”)

p
Solution. 1) We put on Hdiv (Ω) the norm
N
X
k~ukdiv := kui kLp (Ω) + kdiv ~ukLp (Ω)
i=1

p
We can easily check that (Hdiv (Ω), k · kdiv ) is a normed vector space.
We want to prove now that it is complete.
p
Let (~un ) be a Cauchy sequence in Hdiv (Ω).

∀ > 0, ∃n0 > 0, ∀n, p > n0 , k~un − ~up kdiv 6 

By definition of k · kdiv and because (Lp , k · kLp ) is complete, we have

Lp
un,i −−−→ ui ∀i ∈ {1, · · · , N }
n→∞

and
Lp
div ~un −−−→ v
n→∞

18
 Then we have by weak convergence,
D0
un,i * ui weakly ∀i ∈ {1, · · · , N }
n→∞

and so
D0
Di un,i * Di ui weakly ∀i ∈ {1, · · · , N }
n→∞

It follows that
D0
div ~un * div ~u weakly
n→∞

Lp
By limit uniqueness and div ~un −−−→ v, we have v = div ~u.
n→∞
Hence, we have
Lp
un,i −−−→ ui ∀i ∈ {1, · · · , N }
n→∞

and
Lp
div ~un −−−→ div ~u
n→∞
p p
which prove that ~un goes to ~u in Hdiv (Ω) and that ~u ∈ Hdiv (Ω).

2) I am sorry but I didn’t find/take the time to thing about that question...

1.5 Sobolev compact embeddings

Exercise 10. Let Ω bounded of class C 1 , 1 6 p < +∞.

Show that Z Z 1/p
p p
N : u 7→ |∇u| + |tr(u)|
Ω Γ

is a norm over W 1,p (Ω), equivalent to k.k1,p,Ω .

19
Solution. Step 1 : We want to show that N is a norm over W 1,p (Ω).

• absolutely homogeneous: For all u ∈ W 1,p (Ω), for all λ ∈ R,

Z Z 1/p
p p
N (λu) = |∇(λu)| + |tr(λu)|
Ω Γ
Z Z 1/p
p p p p
= |λ| |∇(u)| + |λ| |tr(u)|
Ω Γ
 Z Z 1/p
= |λ|p |∇(u)|p + |tr(u)|p
Ω Γ
Z Z 1/p
p p
= |λ| |∇(u)| + |tr(u)|
Ω Γ
= |λ|N (u)

• point-separating: If u = 0 then N (u) = 0. Conversely, for all u ∈ W 1,p (Ω) such that N (u) = 0.
Then, we have Z Z
p p
k∇ukLp = |∇u| = 0 and |tr(u)|p = 0
Ω Γ

Therefore, we have tr(u(x)) = 0 for all x ∈ Γ, so u ∈ W01,p (Ω).

Then by Poincaré inequality, we have

kukLp 6 Ck∇ukLp

So we have kukLp = 0, and u = 0 because k.kLp is a norm.

• triangle inequality: For all u, v ∈ W 1,p (Ω),

 1
N (u + v) = k∇(u + v)kpLp (Ω) + ktr(u + v)kpLp (Γ)
p

p
p
1
6 k∇ukLp (Ω) + k∇vkLp (Ω) + ktr(u)kLp (Γ) + ktr(v)kLp (Γ) p

We have just used the triangle inequality for k.kLp .

 1  1
N (u + v) 6 k∇ukpLp (Ω) + kvkpLp (Ω) + ktr(u)kpLp (Γ) + ktr(v)kpLp (Γ)
p p

6 N (u) + N (v)

We used the Minkowski discrete inequality.

!1 !1 !1
X p X p X p
p p p
|xi + yi | 6 |xi | + |yi |
i i i

For x1 = k∇ukLp , x2 = ktr(u)kLp and y1 = k∇vkLp , y2 = ktr(v)kLp

Therefore, N is well a norm over W 1,p (Ω).

20
Step 2 : We want to show that there exists C1 such that N (·) 6 C1 k · k1,p,Ω .
We will use the continuity of the trace,

i.e. ∀u ∈ W 1,p (Ω), ktr(u)kLp (Γ) 6 Ckuk1,p,Ω

Then we can prove the wanted inequality. For all u ∈ W 1,p (Ω),

N (u)p = k∇ukpLp (Ω) + ktr(u)kpLp (Γ)

6 k∇ukpLp (Ω) + C p kukp1,p,Ω)
6 kukp1,p,Ω + C p kukp1,p,Ω)
6 C 0 kukp1,p,Ω

Then, by passing to the power p1 , we have the result.

Step 3 : We want to show that there exists C2 such that k · k1,p,Ω 6 C2 N (·).
We work for a contradiction. Suppose that we have for all n ∈ N, un ∈ W 1,p (Ω) such that


kun kLp (Ω) + k∇un kLp (Ω) > n k∇un kLp (Ω) + ktr(un )kLp (Γ) (11)
un
Let consider vn = . So we have
kun k1,p,Ω

kvn k1,p,Ω = 1 (12)

The inequality (11) becomes

1
k∇vn kLp (Ω) + ktr(vn )kLp (Γ) 6
n
Then we have
Lp (Ω) Lp (Γ)
∇vn −−−−→ 0 and tr(vn ) −−−→ 0 (13)
n→∞ n→∞

We use the Poincaré Wirtinger inequality to have

kvn − vn kLp (Ω) 6 Ck∇vn kLp (Ω) −−−→ 0

n→∞
Z
1
where vn = vn . We can remark that
|Ω| Ω

1 C C
|vn | 6 kvn kL1 (Ω) 6 kvn kLp (Ω) 6 because (12).
|Ω| |Ω| |Ω|

Hence, (vn ) is bounded and we can extract a subsequence, still called (vn ), that goes to a constant c.
Lp (Ω) C
By the dominated convergence theorem, we have vn −−−−→ c because Ω is bounded then |vn | 6
n→∞ |Ω|
that is L1 (Ω).
Lp (Ω)
But vn − vn −−−−→ 0, hence
n→∞
Lp (Ω)
vn −−−−→ c
n→∞

21
 By the continuity of the trace, we have
Lp (Γ)
tr(vn ) −−−→ tr(c) = cσ(Γ) = 0 because of (13)
n→∞

where σ(Γ) est the measure of Γ. It follows that c = 0.

We conclude because
Lp (Ω) Lp (Ω)
vn −−−−→ 0 and ∇vn −−−−→ 0
n→∞ n→∞
So
W 1,p (Ω)
vn −−−−−→ 0
n→∞

It is a contradiction with (12).

/ L∞ (RN ).
Exercise 11. Find an example of u ∈ W 1,N (RN ) such that u ∈

Solution. Let take  α

1
u(x) = ln
|x|
/ L∞ . We want to find a condition on
. We can notice that u is not bounded around zero, then u ∈
α such as u(x) ∈ W 1,N (BRN (0, 12 )).
Step 1 : We want to show that u ∈ LN (BRN (0, 21 )).

1 αN
Z Z  
N
|u| = ln dx
B(0,1/2) B(0,1/2) |x|
Z 1/2  αN
1
=K ln ρN −1 dρ as the change of variable in Exercise 5 Step 1
0 ρ
Z +∞
=K xαN e−x(N −1) e−x dx by the change of variable ρ = e−x
ln(2)
Z +∞
αN −xN
=K x
| {ze } dx < +∞
ln(2)
1
!
=O
x2

Then we have the result that u ∈ LN (BRN (0, 21 )).

∂
Step 2 : We want to show that u ∈ LN (BRN (0, 12 )) for all i ∈ J1, N K.
∂xi
Let take i ∈ J1, N K.
∂
We start to find u. For the same reasons of the Exercise 4 Step 3, the weak derivative is
∂xi
equal to the classical derivative.
 α  α−1
∂ 1 1 −1 2xi
ln = α ln |x|
∂xi |x| |x| 2 |x|3
 α−1
1 xi
= −α ln
|x| |x|2

22
 N N
1 α−1 xi
Z Z  
∂
u = α ln dx
B(0,1/2) ∂xi B(0,1/2) |x| |x|2
Z  N (α−1)
N 1 1
6 α ln N
dx
B(0,1/2) |x| |x|
Z 1/2  N (α−1)
1 1 N −1
6K ln ρ dρ
0 ρ ρN
Z +∞
1
6K xN (α−1) −x e−x dx
ln(2) e
Z +∞
6K xN (α−1) dx
ln(2)

And x 7→ xN (α−1) is integrable on [ln(2), +∞[ if and only if N (α − 1) < −1. In other words, we
1
take α < 1 − .
N
1
It follows that, for α < 1 − , we have u ∈ W 1,N (B(0, 1/2)).
N

Step 3 : We want to extend u to be in W 1,N (RN ).

Let take a function ϕ ∈ C0∞ (RN ) such that the support of ϕ is in B(0, 1/2) and ϕ = 1 on a
neighboorhood of zero.

Then uϕ is in W 1,N (RN ) because

u ∈ W 1,N (B(0, 1/2)) and supp(ϕ) ⊂ B(0, 1/2).

And uϕ is not in L∞ (RN ) since the problem of u in zero.

2 Elliptic problems
2.1 Linear problems

Exercise 12. Prove the following inequality :

Z Z
1 2
2
u(x) dx 6 4 k∇uk2 dx
RN kxk RN

for all N > 3 and for all u ∈ H 1 (RN )

Remark : v
u n 2
uX
N
In that exercise, the norm k · k denotes the euclidean norm on R , i.e. kxk = t xi .
i=1

23
Solution. Step 1 : Let prove the result for u ∈ C0∞ (RN ). Let take u ∈ C0∞ (RN ).
We can write Z +∞
2
u (x) = − 2u(tx)∇u(tx).xdt
1

because an antiderivative of 2u(tx)∇u(tx).x is u2 (tx) and when t → +∞, u(tx) → 0 since u ∈

C0∞ (RN ).
Then, we have
Z
1 2
2
u (x)dx
RN kxk
Z Z +∞
2
= 2
u(tx)∇u(tx).xdt dx
RN kxk 1
Z Z +∞
2
6 2
|u(tx)|k∇u(tx)kkxkdtdx by Cauchy Schwarz
RN kxk 1
Z +∞
|u(tx)|
Z
6 2 k∇u(tx)kdtdx
R N 1 kxk
Z +∞
|u(tx)|
Z
6 2 t k∇u(tx)kdtdx because t > 0
RN 1 ktxk
Z +∞ Z
|u(tx)|
6 2t k∇u(tx)kdxdt by Fubini
1 RN ktxk
Z +∞ Z
|u(y)| 1
6 2t k∇u(y)k N dydt by the change of variable y = tx
1 R N kyk t
Z +∞ 1/2 Z 1/2
|u(y)|2
Z
2 2
6 dt dy k∇u(y)k dy by Cauchy-Schwarz
1 tN −1 RN kyk
2
RN

It follows that
1/2 +∞ 1/2
|u(y)|2
Z Z  Z
2 2
dy 6 dt k∇u(y)k dy
RN kyk2 1 tN −1 RN
Z +∞  +∞
2 2 2
And dt = = 6 2 because n > 3.
1 tN −1 (2 − N )tN −2
1 N −2
Finally, while passing to the square, we find

|u(y)|2
Z Z
2
dy 6 4 k∇u(y)k2 dy
R N kyk RN

Step 2 : The general case.

By density of C0∞ (RN ) in H 1 (RN ) for the H 1 −norm, for all u ∈ H 1 (RN ), there exists a sequence
(un )n ∈ C0∞ (RN )N such that
H 1 (RN )
un −−−−−→ u
n→∞

In other words,
L2 (RN ) L2 (RN )
un −−−−−→ u and ∇un −−−−−→ ∇u
n→∞ n→∞

24
 We have, for all n ∈ N,
Z Z
1
un (x)2 dx 6 4 k∇un (x)k2 dx (14)
RN kxk2 RN
| {z }
=k∇un k2 2
L (RN )

We use Fatou’s lemma (because the functions we use are non negative) to have
Z Z
1 2 1
2
u( x) dx = lim inf un (x)2 dx
RN kxk RN kxk2
Z
1
6 lim inf u (x)2 dx
2 n
R N kxk
6 4 lim inf k∇un k2L2 (RN ) by (14)
6 4k∇uk2L2 (RN )
Z
64 k∇u(x)k2 dx
RN

So we have the result for all u ∈ H 1 (RN ).

1
Exercise 13. Let n = 2, 2 < α < 1. Let
n πo
Ωα = (r, θ), 0 < r < 1, 0 < θ <
α
and
u(r, θ) = (r−α − rα ) sin(αθ)

1) Show that ∃q ∗ > 1 such that ∀q ∈ [1, q ∗ [, u ∈ W 1,q (Ωα ).

2) Calculate −∆u

3) What’s up ?

25
 y

Ωα

π/α
x
O 1

2
Solution. 1) Step 1 : We want to show that u ∈ Lq (Ωα ) for q < .
α

π q
Z Z Z 1
q
α
q 1 α
|u| = | sin(αθ)| dθ −r r dr
Ωα 0 0 rα
| {z } | {z }
<∞ 1
∼
r∼0 r αq−1

1 2
The function r 7→ is integrable on ]0, 1[ if and only if αq − 1 < 1, i.e. q < .
rαq−1 α
2
Hence, u ∈ Lq (Ωα ) for q < .
α

∂ ∂ 2
Step 2 : We want to show that u, u ∈ Lq (Ωα ) for q < .
∂x ∂y α+1
While deriving u, we have
∂
u = sin(αθ)(−αr−α−1 − αrα−1 ) (15)
∂r
∂
u = α cos(αθ)(r−α − rα ) (16)
∂θ

But
∂ ∂ ∂

 u = cos(θ) u + sin(θ) u

∂r ∂x y
1 ∂ ∂ ∂

 u = − sin(θ) u + cos(θ) u
r ∂θ ∂x ∂y
∂ 1 ∂ ∂ ∂
Hence, if u and u are in Lq , then u, u will be in Lq .
∂r r ∂θ ∂x ∂y

26
 •
q π q
Z Z Z 1
∂ α 1
u = |α sin(αθ)|q dθ + rα−1 r dr
Ωα ∂r rα+1
|0 {z } 0
| {z }
<∞ 1
∼
r∼0 r(α+1)q−1

1
The function r 7→ (α+1)q−1 is integrable on ]0, 1[ if and only if (α + 1)q − 1 < 1, i.e.
r
2
q< .
α+1
•
q π q
Z Z Z 1
1 ∂ α
q 1 1 α
u = |α cos(αθ)| dθ − r r dr
Ωα r ∂θ rq rα
|0 {z } 0
| {z }
<∞ 1
∼
r∼0 r (α+1)q−1

1
The function r 7→ (α+1)q−1 is integrable on ]0, 1[ if and only if (α + 1)q − 1 < 1, i.e.
r
2
q< .
α+1
∂ ∂ 2
It follows that u, u ∈ Lq (Ωα ) for q < .
∂x ∂y α+1
2
We take q ∗ = to have u ∈ W 1,q (Ωα ) for all q ∈ [1, q ∗ [.
α+1

∂2 1 ∂ 1 ∂2
2) We have ∆u = u + u + u.
∂r2 r ∂r r2 ∂θ2
And we can compute

∂2 −α−2 α−2


u = sin(αθ α(α + 1)r − α(α − 1)r while deriving (15)
∂r2
1 ∂
u = sin(αθ)(−αr−α−2 − αrα−2 ) while using (15)
r ∂r
1 ∂2
u = −α2 sin(αθ)(r−α−2 − rα−2 ) while deriving (16)
r2 ∂θ2

While computing the laplacian, we find that ∆u = 0.

3) It follows that the PDE (

−∆v = 0
v|∂Ωα = 0

has two solutions in W01,q (Ωα ) : u and the zero function.

The problem is that Ωα is not of class C 1 because at the point (0, 0) we have an angle. Then
we lose the uniqueness of the solution.

27