Master 2 Maths - Basic course A5: An introduction to elliptic

PDEs and evolution problems

Mihai MARI“

Institut de Mathématiques de Toulouse

e-mail: [email protected].

1 Crash course: functional analysis, distributions, Sobolev spaces

1.1 Principles of functional analysis

Recommended reading: [2] or [3], Chapters I-II.
Theorem 1. (Hahn-Banach) Let E be a vector space and p : E −→ R be a function satisfying
p(λx) = λp(x) for any x ∈ E and any λ > 0,
p(x + y) ≤ p(x) + p(y) for all x, y ∈ E.
Let G ⊂ E be a linear subspace and let g : G −→ R be a linear functional satisfying
g(x) ≤ p(x) for any x ∈ G.
Then there exists a linear functional φ : E −→ R that extends g , that is φ(x) = g(x) for all
x ∈ G, and satises
φ(x) ≤ p(x) for any x ∈ E.
Let (E, k · k) be a normed vector space. We denote by E ∗ its dual space, namely
E ∗ = {φ : E −→ R | φ is linear and continuous }

endowed with the norm

kφkE ∗ = sup |φ(x)|.
x∈E, kxk61

We recall that (E ∗ , k · kE ∗ ) is a Banach space (i.e. it is complete, which means that any Cauchy
sequence is convergent) even when E is not complete.
Corollary 2. Let E be a normed vector space and G ⊂ E be a linear subspace. If g : G −→ R
is a continuous functional, there exists φ ∈ E ∗ that extends φ (i.e. φ(x) = g(x) for all x ∈ G)
and satises
kφkE ∗ = kgkG∗ .

Corollary 3. Let E be a normed vector space. For any x0 ∈ E there exists φ0 ∈ E ∗ such that
kφ0 kE ∗ = kx0 k and φ0 (x0 ) = kx0 k2 .

Proof. Use the previous Corollary with G = Rx0 and g(tx0 ) = tkx0 k2 .
Denition 4. Let E be a vector space. An ane hyperplane in E is a set of the form
Hφ,α = {x ∈ E | φ(x) = α}, where φ : E −→ R is a linear mapping (not necessarily
continuous), φ 6≡ 0 and α ∈ R.

1
Proposition 5. The ane hyperplane Hφ,α is closed if and only if (i) φ is continuous.
Denition 6. Assume that E is a vector space, A, B ⊂ E and Hφ,α is an ane hyperplane.
We say that Hφ,α separates A and B if
φ(x) ≤ α for any x ∈ A and φ(x) ≥ α for any x ∈ B.
We say that Hφ,α strictly separates A and B if there exists ε > 0 such that
φ(x) ≤ α − ε for any x ∈ A and φ(x) ≥ α + ε for any x ∈ B.

Theorem 7. (Hahn-Banach, rst geometric form) Let (E, k · k) be a normed vector space. Let
A ⊂ E and B ⊂ E be two nonempty convex sets such that A ∩ B = ∅. Assume that one of
them is open. Then there exists a closed ane hyperplane that separates A and B .
Theorem 8. (Hahn-Banach, second geometric form) Let (E, k · k) be a normed vector space.
Let A ⊂ E and B ⊂ E be two nonempty convex sets such that A ∩ B = ∅. Assume that
A is closed and B is compact. Then there exists a closed ane hyperplane that strictly
separates A and B .
The next results rely on the completeness of the subsequent Banach spaces.
Theorem 9. (uniform boundedness principle; Banach-Steinhaus) Let E and F be two Banach
spaces. Let (Ti )i∈I be a family of linear continuous operators from E to F (I is not necessarily
countable). Assume that
sup kTi xkF < ∞ for any x ∈ E.
i∈I

Then
sup kTi kL(E,F ) < ∞.
i∈I
In other words, there exists C > 0 such that
kTi xkF ≤ CkxkE for any x ∈ E and any i ∈ I.

Corollary 10. Let E and F be two Banach spaces. Let (Tn )n≥1 be a sequence of linear
continuous operators from E to F such that for every x ∈ E , the sequence (Tn x)n≥1 converges
in F to a limit that we denote T x. Then:
(i) supn≥1 kTn kL(E,F ) < ∞,
(ii) T ∈ L(E, F ),
(iii) kT kL(E,F ) ≤ lim inf n→∞ kTn kL(E,F ) .
Corollary 11. Let E be a Banach space and let B ⊂ E . Assume that for any φ ∈ E ∗ the set
{φ(x) | x ∈ B} is bounded. Then B is bounded in E .
Theorem 12. (open mapping theorem) Let E and F be two Banach spaces. Assume that
T : E −→ F is a linear operator that is surjective (i.e. T (E) = F ). Then there exists c > 0
such that BF (0, c) ⊂ T (BE (0, 1)).
Corollary 13. Let E and F be two Banach spaces. Assume that T : E −→ F is a linear
operator that is bijective. Then T −1 : F −→ E is continuous.
Theorem 14. (closed graph theorem) Let E and F be two Banach spaces. Assume that
T : E −→ F is a linear operator. Then T is continuous if and only if the graph of T ,
G(T ) = {(x, T x) | x ∈ E},
is closed in E × F.

2
1.2 Weak topologies. Reexive spaces. Separable spaces.
See Chapter III in [2] or [3].
Lemma 1. Let X be a set and (Ui )i∈I a family of subsets of X . The smallest topology T on
X containing all Ui 's is the family of arbitrary unions of nite intersections of the sets Ui :
   
 [ \ 
T =  Ui  i ∈ I .
arbitrary nite
 

The topology T above is called the topology generated by the family (Ui )i∈I .

Lemma 2. Assume that X is a set, (Yi , Ti )i∈I is a collection of topological spaces and for each
i ∈ I , φi is a maping from X to Yi .
(i) The smallest topology on X with
[ the property that each mapping φi is continuous is the
topology T generated by the family φ−1
i (O) O ∈ Ti .
i∈I

(ii) Let (xn )n≥1 be a sequence in X . Then xn −→ x (in T ) if and only if φi (xn ) −→ φi (x)
in (Yi , Ti ) for any i ∈ I .
(iii) Let Z be an arbitrary topological space and let ψ : Z −→ X , where X is endowed with
the topology T . Then ψ is continuous if and only if φi ◦ ψ is continuous from Z to Yi for every
i ∈ I.

Let E be a Banach space and let E ∗ be its dual space. Now we ignore the usual topology
on E (associated with the norm k · k) and we dene a new topology as follows:
Denition 3. The weak topology σ(E, E ∗ ) is the smallest topology on E such that all map-
pings f ∈ E ∗ are continuous from E to R.
Since every f ∈ E ∗ is continuous for the usual topology (also called the strong topology), it
follows that the weak topology is smaller (weaker) than the usual topology.
Notation. If (xn )n≥1 converges to x in the weak topology σ(E, E ∗ ) we write
xn * x.

Proposition 4. The weak topology σ(E, E ∗ ) is Hausdor.

Proposition 5. Let (xn )n≥1 be a sequence in E . We have:
(i) xn * x in σ(E, E ∗ ) ⇔ f (xn ) −→ f (x) for any f ∈ E ∗ .
(ii) If xn −→ x strongly, then xn * x weakly in σ(E, E ∗ )
(iii) If xn * x weakly in σ(E, E ∗ ), then (kxn k)n≥1 is bounded and kxk ≤ lim inf n→∞ kxn k.
(iv) If xn * x weakly in σ(E, E ∗ ) and fn −→ f strongy in E ∗ then fn (xn ) −→ f (x).
Proposition 6. If E is nite-dimensional, the weak and the strong topoologies on E are the
same.
Example 7. Let E be an innite dimensional Banach space. Then the closure of the unit
sphere S = {x ∈ E | kxk = 1} in the weak topology σ(E, E ∗ ) is B(0, 1) = {x ∈ E | kxk ≤ 1}.
Thus the weak and the strong topologies are never identical in innite dimensional Banach
spaces.
Theorem 8. Let E be a Banach space and let C be a convex subset of E . Then C is closed
in the weak topology σ(E, E ∗ ) if and only if C is closed in the strong topology.

3
Corollary 9. (Mazur) Assume that (xn )n≥1 converges weakly to x. Then there exists a
sequence (yn )n≥1 made up of linear combinations of the xn 's that converges strongly to x.
Theorem 10. Let E and F be two Banach spaces and let T : E −→ F be a linear operator.
Then T is continuous from E to F (endowed with the strong topologies) if and only if T is
continuous from E endowed with σ(E, E ∗ ) to F endowed with σ(F, F ∗ ).

Exercise 11. Let E be a Banach space. For each x ∈ E we dene a mapping J(x) : E ∗ −→ R
by hJ(x), f i = f (x) for any f ∈ E ∗ .
a) Prove that for any x ∈ E the mapping J(x) is linear, continuous on E ∗ (thus belongs to
E ∗∗ ) and kJ(x)kE ∗∗ = kxkE .
b) Prove that J : E −→ E ∗∗ is a linear isometry from E to the linear subspace J(E) ⊂ E ∗∗ .
(This enables us to identify E to the subspace J(E) ⊂ E ∗∗ .)
Denition 12. The mapping J in the previous exercise is called the canonical injection from
E to E ∗∗ .
Denition 13. Let E be a Banach space and let E ∗ be its dual space. The weak∗ topology
on E ∗ is the smallest topology on E ∗ with the property that for any x ∈ E , the mapping J(x)
is continuous on E ∗ . We denote by σ(E ∗ , E) the weak∗ topology on E ∗ .
A sequence (fn )n≥1 ⊂ E ∗ converges to f ∈ E ∗ in the weak∗ topology if and only if fn (x) −→
∗
f (x) for any x ∈ E . We write fn * f.
So far we have introduced three topologies on E ∗ : the usual (strong) topology induced by
the norm k · kE ∗ , the weak topology σ(E ∗ , E ∗∗ ) and the weak∗ topology σ(E ∗ , E). The weak∗
topology σ(E ∗ , E) is weaker than the weak topology σ(E ∗ , E ∗∗ ) which is weaker than the usual
topology.
Proposition 14. The weak∗ topology σ(E ∗ , E) is Hausdor.
Proposition 15. Let E be a Banach space and let (fn )n≥1 be a sequence in E ∗ . We have:
∗
(i) fn * f in σ(E ∗ , E) ⇔ fn (x) −→ f (x) for any x ∈ E.
(ii) If fn −→ f strongly, then fn * f weakly in σ(E ∗ , E ∗∗ ).
∗
If fn * f weakly in σ(E ∗ , E ∗∗ ), then fn * f in σ(E ∗ , E).
∗
(iii) If fn * f in σ(E ∗ , E), then (kfn k)n≥1 is bounded and kf k ≤ lim inf kfn k.
n→∞
∗
(iv) If fn * f in σ(E ∗ , E) and xn −→ x strongy in E then fn (xn ) −→ f (x).
Theorem 16. (Banach-Alaoglu-Bourbaki) Let E be an arbitrary Banach space. The closed
unit ball B E ∗ (0, 1) = {f ∈ E ∗ | kf kE ∗ ≤ 1} is compact in the weak∗ topology σ(E ∗ , E).
Denition 17. Let E be a Banach space and let J : E −→ E ∗∗ be the canonical injection
from E to The space E is called reexive if J is surjective, that is J(E) = E ∗∗ .
E ∗∗ .
When E is reexive we usually identify E with E ∗∗ .
Theorem 18. (Kakutani) A Banach space E is reexive if and only if B E (0, 1) = {x ∈
E | kxkE ≤ 1} is compact in the weak topology σ(E, E ∗ ).
Theorem 19. Assume that E is a reexive Banach space and (xn )n≥1 is a bounded sequence
in E . Then there exists a subsequence (xnk )k≥1 that converges in the weak topology σ(E, E ∗ ).
We mention that the converse of the above theorem is true: a Banach space E with the
property that any bounded sequence in E admits a weakly convergent sequence in σ(E, E ∗ ). is
reexive. This result is due to Eberlein and Šmulian.
Theorem 20. Asume that E is a reexive Banach space and G ⊂ E is a closed linear subspace
of E . Then G is reexive.

4
Corollary 21. A Banach space E is reexive if and only if its dual space E ∗ is reexive.
Corollary 22. Let E be a reexive Banach space. Let K ⊂ E be a bounded, closed and
convex subset of E . Then K is compact in the weak topology σ(E, E ∗ ).
Exercise 23. Let H be a Hilbert space that has a countable Hilbert basis B = (en )n≥1 . Let
S(0, 1) = {x ∈ H | kxk = 1}.
i) Prove that en * 0 in H as n −→ ∞.
ii) Is the set S(0, 1) closed and bounded in H ? Is it compact in the weak topology σ(H, H ∗ )?
iii) Let x ∈ B(0, 1). Prove that there exists a sequence (xn )n≥1 ⊂ S(0, 1) such that xn * x
in H as n −→ ∞.
iv) Is the set B(0, 1) sequentially compact in the weak topology σ(H, H ∗ )? Is it compact?
Corollary 24. Let E be a reexive Banach space and let A ⊂ E , A 6= ∅ be a closed convex
set. Assume that Φ : A −→ (−∞, ∞] is a convex lower-semicontinuous function (that is, for
any λ ∈ R the set {x ∈ A | Φ(x) ≤ λ} is closed) such that Φ 6≡ ∞ and
lim Φ(x) = ∞ (no assumption if A is bounded).
x∈A, kxk→∞

Then Φ achieves its minimum on A, i.e. there exists x0 ∈ A such that Φ(x0 ) = minx∈A Φ(x).

Denition 25. A metric space (X, d) is separable if there exists a subset D ⊂ X that is
countable and dense.
Proposition 26. Assume that (X, d) is a separable metric space. Let Y be any subset of X .
Then (Y, d) is also separable.
Theorem 27. Let E be a Banach space such that E ∗ is separable. Then E is separable.
Corollary 28. Let E be a Banach space. Then
E is reexive and separable ⇔ E ∗ is reexive and separable.
Theorem 29. Let E be a Banach space. Then E is separable if and only if B E ∗ (0, 1) is
metrizable in the weak∗ topology σ(E ∗ , E).
Theorem 30. Let E be a Banach space. Then E ∗ is separable if and only if B E (0, 1) is
metrizable in the weak topology σ(E, E ∗ ).
Theorem 31. Let E be a separable Banach space. Then any bounded sequence (fn )n≥1 ⊂ E ∗
possesses a subsequence (fnk )k≥1 that converges in the weak∗ topology σ(E ∗ , E).

Denition 32. A Banach space is said to be uniformly convex if for any ε > 0 there exists
δ > 0 such that
x+y
∀x ∈ E, ∀y ∈ E, kxk ≤ 1, kyk ≤ 1 and kx − yk > ε ⇒ < 1 − δ.
2

Example 33.
1
RN endowed with the norm kxkp = (|x1 |p + · · · + |xN |p ) p is uniformly convex.
RN endowed with the norm k · k1 or kxk∞ = max1≤j≤N |xj | is not uniformly convex.
Theorem 34. (Milman-Pettis) Any uniformly convex Banach space is reexive.
Theorem 35. Assume that E is a uniformly convex Banach space. Let (xn )n≥1 ⊂ E be a
sequence such that xn * x weakly in σ(E, E ∗ ) and kxn k −→ kxk as n −→ ∞. Then xn −→ x
strongly in E .

5
1.3 Lp spaces
References: Chapter IV in [2] or [3].
Let Ω ⊂ RN be an open set. We identify functions that are equal almost everywhere (a.e.) on Ω.
Denition 1. Let 1 ≤ p < ∞. We dene
Z
L (Ω) = {f : Ω −→ [−∞, ∞] | f is Lebesgue measurable and
p
|f (x)|p dx < ∞}.
Ω

Z 1
p
We denote kf kp = |f (x)| dxp
.
Ω
Denition 2.
L∞ (Ω) = {f : Ω −→ [−∞, ∞] | f is Lebesgue measurable
and there exists C ≥ 0 s. t. |f | ≤ C a.e.}.
We dene kf k∞ = inf{C ≥ 0 |f | ≤ C a.e.}.
Theorem 3. (Hölder's inequality) Let 1 ≤ p ≤ ∞. Let p0 be the conjugate exponent of p,
namely = 1. Then for any f ∈ Lp (Ω) and any g ∈ Lp (Ω) we have f g ∈ L1 (Ω) and
1 1 0
p + p0
Z
|f g| dx ≤ kf kp kgkp0 .
Ω

Exercise 4. Prove the generalized Hölder inequality: if f1 ∈ Lp1 (Ω), . . . , fn ∈ Lpn (Ω), where
p1 , . . . , pn ∈ [1, ∞] and 1
p1 + ··· + 1
pn = 1, then f1 f2 . . . fn ∈ L1 (Ω) and
Z
|f1 f2 . . . fn | dx ≤ kf1 kp1 kf2 kp2 . . . kfn kpn .
Ω

Theorem 5. For any 1 ≤ p ≤ ∞, k · kp is a norm on Lp (Ω) and Lp (Ω) endowed with this
norm is a Banach space.
Theorem 6. (density of smooth functions) C0∞ (Ω) is dense in Lp (Ω) for 1 ≤ p < ∞.
Theorem 7. Lp (Ω) is separable for any 1 ≤ p < ∞.
Remark 8. L∞ (Ω) is not separable. C0∞ (Ω) is not dense in L∞ (Ω).
Theorem 9. Lp (Ω) is uniformly convex (and consequently it is reexive) for any 1 < p < ∞.
Remark 10. L1 (Ω) and L∞ (Ω) are not reexive.
Theorem 11. (Riesz) Let 1 ≤ p < ∞. For any ϕ ∈ (Lp (Ω))∗ there exists g ∈ Lp0 (Ω) such that
Z
ϕ(f ) = g(x)f (x) dx for any f ∈ Lp (Ω).
Ω

Moreover, g is unique (remember that we identify functions that are equal a.e.) and the
mapping ϕ 7−→ g is a linear isometry from (Lp (Ω))∗ to Lq (Ω).
The isometry ϕ 7−→ g enables us to identify the dual of Lp (Ω) with Lp (Ω).
0

6
1.4 Distributions
In this section we closely follow [1], chapters 4-7. Students not familiar with French may nd
an elementary presentation with exercises in [7]. One can also consult Chapter 6 in [6].
Denition 1. Let Ω ⊂ RN be an open set. A distribution in Ω is a linear form on C0∞ (Ω)
satisfying the following continuity property: for any compact set K ⊂ Ω there exist p ∈ N and
C > 0 such that

∀ϕ ∈ C0∞ (Ω) with supp(ϕ) ⊂ K there holds |hu, ϕi| 6 C sup |∂ α ϕ(x)|.
x∈K,|α|6p

We denote by D0 (Ω) the vector space of distributions in Ω.

If the integer p in the above inequality can be chosen independently of K we say that u is
a distribution of nite order; the smallest integer with this property is the order of u.
Denition 2. A sequence of distributions (un )n>1 ⊂ D0 (Ω) converges to u ∈ D0 (Ω) i
hun , ϕi −→ hu, ϕi for any ϕ ∈ C0∞ (Ω). We write un −→ u in D0 (Ω).
Theorem 3. Let (un )n>1 ⊂ D0 (Ω) be a sequence of distributions. Assume that for any
ϕ ∈ C0∞ (Ω) the sequence (hun , ϕi)n>1 has a limit that we denote hu, ϕi. Then u is a distribution
in Ω.
Example 4. a) Let f ∈ L1loc (Ω). Then f denes a distribution (usually denoted Tf or simply
f ) given by Z
hTf , ϕi = hf, ϕi = f (x)ϕ(x) dx.
Ω
It is clear that for any compact set K ⊂ Ω and for any ϕ ∈ C0∞ (Ω) with supp(ϕ) ⊂ K we have
Z  
|hTf , ϕi| 6 |f (x)| dx sup |ϕ(x)| ,
K x∈K

hence Tf is a distribution of order 0.

If f, g ∈ L1loc (Ω) and Tf = Tg in D0 (Ω), then f = g a.e. in Ω.
In the sequel we identify Tf with f and write hf, ϕi instead of hTf , ϕi.
If fn −→ f in L1loc (Ω) then fn −→ f in D0 (Ω) (in particular, this happens when fn −→ f
a.e. and there exists g ∈ L1loc (Ω) such that |fn | 6 g a.e.).
b) The Dirac mass at a ∈ RN is dened by hδa , ϕi = ϕ(a). It is a distribution of order 0.
c) Let a ∈ RN and α ∈ NN . The linear form dened by hu, ϕi = ∂ α u(a) is a distribution
of order |α|.
Exercise 5. (principal value) We dene the linear form vp 1
on C0∞ (R) by


x
Z
ϕ(x)
hvp 1


x , ϕi = lim dx.
ε→0 {|x|>ε} x

a) Prove that vp x is well-dened and is

1
a distribution of order 6 1.

b) 1. (Hint: consider its action on the functions

Prove that vp x1 is precisely of order

ϕn (x) where
= χ(x) arctan(nx) is even and equals 1 in a neighborhood of 0.)
χ ∈ C0∞ (R)
Theorem 6. Assume that u ∈ D0 (Ω) is a positive distribution, which means that hu, φi > 0
whenever φ ∈ C0∞ (Ω) and φ > 0. Then u is of order 0 (in fact, u is a positive Radon measure).
Theorem 7. Assume that the sequence of functions (ρn )n>1 ⊂ L1 (RN ) satises the following
properties:

7
 (i) ρn > 0,
Z
(ii) ρn (x) dx −→ 1 as n −→ ∞,
RN
(iii) supp(ρn ) ⊂ B(0, rn ) where rn −→ 0 as n −→ ∞.
Then ρn −→ δ0 in D0 (RN ).
Denition 8. If, in addition to (i), (ii), (iii) in the above Theorem we have ρn ∈ C0∞ (RN ) for
all n, we say that (ρn )n>1 is a sequence of molliers in RN .
Let f ∈ C 1 (Ω). Integrating by parts we get for any ϕ ∈ C0∞ (Ω),
D ∂f E Z ∂f
Z
∂ϕ D ∂ϕ E
,ϕ = (x)ϕ(x) dx = − f (x) (x) dx = − f, .
∂xi Ω ∂xi Ω ∂xi ∂xi
This justies the following
∂u
Denition 9. For any u ∈ D0 (Ω) we dene by
∂xi
D ∂u E D ∂ϕ E
, ϕ = − u, .
∂xi ∂xi

∂u
Exercise 10. Check that is well-dened and is a distribution in D0 (Ω). If u is of order k,
∂xi
∂u
then is of order k + 1.
∂xi
Denition 11. For any u ∈ D0 (Ω) and any multiindex α we dene ∂ α u by
h∂ α u, ϕi = (−1)|α| hu, ∂ α ϕi.

Clearly, for any u ∈ D0 (Ω) and for any multiindices α and β we have ∂ α+β u = ∂ α ∂ β u .

Theorem 12. If un −→ u in D0 (Ω) then ∂ α un −→ ∂ α u in D0 (Ω) for any multiindex α.

0 if x < 0

Exercise 13. Consider the Heaviside function H : R −→ R, H(x) =
1 if x > 0.
Prove
that H 0 = δ0 in D0 (R).
Remark 14. Some simple operations with distributions:
- Restriction. Assume that ω ⊂ Ω and ω is open. Given u ∈ D0 (Ω), we dene the
restriction of u to ω by
hu|ω , ϕi = hu, ϕi for any ϕ ∈ C0∞ (ω).

- Complex conjugation. hu, ϕi = hu, ϕi.

- Translation. If f is a function dened on RN and a ∈ RN we dene τa f (x) = f (x − a).
If u ∈ D0 (RN ) we dene τa u by hτa u, ϕi = hu, τ−a ϕi.
- Dilation. Let λ ∈ R, λ 6= 0. If f is a function let fλ (x) = f ( λx ). If u is a distribution in
RN we dene huλ , ϕi = |λ|N hu, ϕ 1 i.
λ
Particular case: λ = −1. We denote fˇ(x) = f (−x) and ǔ = u−1 .
Exercise 15. a) Let u ∈ D0 (R). Prove that 1
− u) −→ u0 in D0 (R) as h −→ 0.
h (τ−h u
b) State and prove a similar result for partial derivatives in D0 (RN ).

8
Exercise 16. a) Let f : R −→ R be a nondecreasing function. Prove that f 0 is a positive
Radon measure, where f 0 is the derivative of f in D0 (R). (Hint: Use the theorem on positive
distributions.)
b) Conversely, let µ be a positive Borel measure on R such that µ(R) < ∞. Dene F : R −→
R by FR(t) = µ((−∞, t]). Prove that F is nondecreasingR and R (Hint: Write
R F = µ in DR (R).
0 0

F (t) = R 1{s6t} dµ(s), then use Fubini's theorem to get R R . . . dµ dx = R R . . . dx dµ.)

Denition 17. Let u ∈ D0 (Ω) and let f ∈ C ∞ (Ω). We dene f u by
hf u, ϕi = hu, f ϕi for any ϕ ∈ C0∞ (Ω).

Exercise 18. Prove that f u is a distribution.

X α 
Exercise 19. Prove Leibniz' formula: ∂ α (f u) =
β
∂ β f · ∂ α−β u.
β6α

Theorem 20. Assume that un −→ u in D0 (Ω),

f, fn ∈ C ∞ (Ω) and ∂ α fn −→ ∂ α f uniformly
on each compact set K ⊂ Ω. Then fn un −→ f u in D0 (Ω).
Surface integrals and Stokes' formula. Let ω ⊂ RN −1 be an open set.
Let Φ : ω −→ (a, b) be a C 1 function. Denote by Γ the graph of Φ, namely
Γ = {x = (x0 , xN ) | x0 = (x1 , . . . ,N −1 ) ∈ ω, xN = Φ(x0 )}.

Recall that:
• Γ is a (simple) surface in RN .
• The unit normal toΓ pointing "upward"
 (i.e. in the direction of OxN ) at a point (x , Φ(x ))
0 0

−∂1 Φ(x0 )
..
1 .
 
is n = p 
.

1 + |∇Φ(x0 )|2  −∂N −1 Φ(x0 ) 


1
• Given an open set ω 0 ⊂ ω , the surface measure of the set
Γ0 := {x = (x0 , xN ) | x0 = (x1 , . . . ,N −1 ) ∈ ω 0 , xN = Φ(x0 )} ⊂ Γ is
Z Z p
1 dσ = 1 + |∇Φ(x0 )|2 dx1 . . . dxN −1 .
Γ0 ω0

• If f : Γ −→ R is continuous and has compact support, the integral of f with respect to

the surface measure σ on Γ is
Z Z
f (x0 , Φ(x0 )) 1 + |∇Φ(x0 )|2 dx1 . . . dxN −1 .
p
f dσ =
Γ ω

We say that an open set Ω ⊂ RN has C 1 boundary i for any point P ∈ ∂Ω there exist:
- an orthonormal system of coordinates (x) = (x1 , . . . , xN )
- an open set ω ⊂ RN −1 such that p0 ∈ ω , where (p1 , . . . , pN ) = (p0 , pN ) are the coordinates
of P in the system of coordinates (x),
- a mapping Φ : ω −→ (a, b) of class C 1 such that, in the system of coordinates (x),
Ω ∩ (ω × (a, b)) = {x = (x0 , xN ) ∈ ω × (a, b) | xN > Φ(x0 )}.

Notice that Ω is, locally, on the same side of ∂Ω.

If Ω is open, bounded, with C 1 boundary, then ∂Ω is compact and for each point P ∈ ∂Ω
there is a "cylinder" as above. We may cover ∂Ω by a nite number of such cylinders. Using
a partition of unity we extend the notion of (outer) unit normal and surface measure to ∂Ω.

9
 A function f dened on Ω belongs to C m (Ω) if there exists a function F ∈ C m (RN ) such
that f = F|Ω .
Theorem 21. (Stokes' formula) Assume that Ω ⊂ RN is open, bounded, with C 1 boundary.
Assume that F = (f1 , . . . , fN ) ∈ C 1 (Ω). Then
Z Z Z
∂f1 ∂fN
div(F ) dx = + ··· + dx = F · n dσ.
Ω Ω ∂x1 ∂xN ∂Ω

Corollary 22. (integration by parts) Assume that Ω ⊂ RN is open, bounded, with C 1

boundary. Let f, g ∈ C 1 (Ω). Then
Z Z Z
∂g ∂f
f (x) dx = f (x)g(x)ni dσ − g(x) dx.
Ω ∂xi ∂Ω Ω ∂xi

Support and convolution of distributions

Denition 23. Let T ∈ D0 (Ω). We dene

F(T ) := {F ⊂ Ω | F is closed in Ω and T|Ω\F = 0} and supp(T ) =

\
F.
F ∈F (T )

In other words, supp(T ) is the smallest closed set F ⊂ Ω such that T|Ω\F = 0.
Proposition 24. (Basic properties) Let T, S ∈ D0 (Ω). The following assertions hold.
a) T|Ω\supp(T ) = 0.
b) For any ϕ ∈ C0∞ (Ω) with supp(T ) ∩ supp(ϕ) = ∅ we have hT, ϕi = 0.
c) supp(S + T ) ⊂ supp(S) ∪ supp(T ).
d) For any a ∈ C ∞ (Ω) we have supp(aT ) ⊂ supp(a) ∩ supp(T ).
e) For any multiindex α ∈ NN we have supp(∂ α T ) ⊂ supp(T ).
Remark 25. If ϕ ∈ C0∞ (Ω) and ϕ = 0 on supp(T ) does not imply hT, ϕi = 0.
Exercise 26. Prove that supp(∂ α (δ0 )) = {0}. Show that there is ϕ ∈ C0∞ (R) such that
ϕ(0) = 0 and hδ00 , ϕi = 0.
Denition 27. The space of distributions with compact support in Ω is
E 0 (Ω) = {T ∈ D0 (Ω) | supp(T ) is a compact subset of Ω}.

Proposition 28. Let T ∈ E 0 (Ω). Assume that χ1 , χ2 ∈ C0∞ (Ω) are equal to 1 in an open
neighborhood of supp(T ). Then for any ϕ ∈ C ∞ (Ω) we have
hT, χ1 ϕi = hT, χ2 ϕi.

Denition 29. Let T ∈ E 0 (Ω). Consider χ ∈ C0∞ (Ω) such that χ = 1 in an open neighborhood
of supp(T ). Then T can be extended to a linear (and continuous - see below) functional on
C ∞ (Ω) dened as follows:

∀ϕ ∈ C ∞ (Ω), hT, ϕiE 0 (Ω),C ∞ (Ω) = hT, χϕiD0 (Ω),C0∞ (Ω) .

Proposition 30. Let T ∈ E 0 (Ω). Then:

10
 a) There exist a compact set K ⊂ Ω, p ∈ N and C > 0 such that

∀ϕ ∈ C ∞ (Ω), hT, ϕiE 0 (Ω),C ∞ (Ω) ≤ C max sup |∂ α ϕ(x)|.

|α|≤p x∈K

b) T is of nite order.
c) Assume that (ϕn )n≥1 ⊂ C ∞ (Ω), ϕ ∈ C ∞ (Ω) and ∂ α ϕn −→ ∂ α ϕ uniformly on any
compact set in Ω. Then hT, ϕn i −→ hT, ϕi.
Denition 31. Let T ∈ E 0 (Ω). Then T can be extended to a distribution T̃ ∈ E 0 (RN ) as
follows:
∀ϕ ∈ C ∞ (RN ), hT̃ , ϕi = hT, ϕ|Ω i.

Convolution C0∞ ∗ D0

Denition 32. Let T ∈ D0 (RN ), ϕ ∈ C0∞ (RN ). Dene T ∗ ϕ by

T ∗ ϕ(x) = hT, ϕ(x − ·)i.

Remark. If T = Tf with f ∈ L1locR (RN ) we recover the standard denition of convolution of

functions: Tf ∗ ϕ(x) = f ∗ ϕ(x) = RN f (y)ϕ(x − y) dy .
Proposition 33. supp(T ∗ ϕ) ⊂ supp(T ) + supp(ϕ).
Proposition 34. T ∗ ϕ ∈ C ∞ (RN ) and ∂ α (T ∗ ϕ) = (∂ α T ) ∗ ϕ = T ∗ (∂ α ϕ).
Denition 35. (Convolution E 0 ∗ C ∞ )
If T ∈ E 0 (RN ) and ϕ ∈ C ∞ (RN ) we dene T ∗ ϕ by the same formula as above:
T ∗ ϕ(x) = hT, ϕ(x − ·)i.

This denition makes sense and, as before, T ∗ ϕ is a C ∞ function in RN .

Theorem 36. (Regularization of distributions) Let T ∈ D0 (RN ). Let (ζn )n≥1 be a sequence
of molliers. Then Tn := T ∗ ζn ∈ C ∞ (RN ) and Tn −→ T in D0 (RN ).
Theorem 37. (Density of C0∞ (Ω) in D0 (Ω)) Consider an open set Ω ⊂ RN and T ∈ D0 (Ω).
There exists a sequence (Tn )n≥1 ⊂ C0∞ (Ω) such that Tn −→ T in D0 (Ω) as n −→ ∞.
Theorem 38. Let ϕ ∈ C0∞ (RN ). Assume that (Tn )n≥1 ⊂ D0 (RN ) and Tn −→ T in D0 (RN ).
Then for any multiindex α ∈ NN we have ∂ α (Tn ∗ ϕ) −→ ∂ α (T ∗ ϕ) uniformly on any compact
set K ⊂ RN .

Convolution of distributions

Remark 39. For any T ∈ D0 (RN ) and ϕ, ψ ∈ C0∞ (RN ) we have hT ∗ ψ, ϕi = hT, ψ̌ ∗ ϕi.
Denition 40. Let T ∈ D0 (RN ) and S ∈ E 0 (RN ). We dene T ∗ S and S ∗ T by
∀ϕ ∈ C0∞ (RN ), hT ∗ S, ϕi = hT, Š ∗ ϕi.

∀ϕ ∈ C0∞ (RN ), hS ∗ T, ϕi = hS, Ť ∗ ϕi.

Exercise 41. Show that T ∗ S and S ∗ T are distributions.

Proposition 42. supp(T ∗ S) ⊂ supp(T ) + supp(S).

11
Theorem 43. (Commutativity) For any T ∈ D0 (RN ) and S ∈ E 0 (RN ) we have T ∗ S = S ∗ T .
Theorem 44. Assume that
i) Tn −→ T in D0 (RN ),
ii) Sn −→ S in D0 (RN ) and there is a compact set K such that supp(Sn ) ⊂ K for all n.
Then Tn ∗ Sn −→ T ∗ S in D0 (RN ) as n −→ ∞.
Example 45. a) For any T ∈ D0 (RN ) we have T ∗ δ0 = T .
b) Let (ζn )n≥1 ⊂ C0∞ (RN ) be a sequence of molliers. We have seen that ζn −→ δ0 in
D (RN ). The previous theorem implies that T ∗ ζn −→ T in D0 (RN ) as n −→ ∞.
0

Theorem 46. Let T ∈ D0 (RN ). Let S ∈ E 0 (RN ). For any multiindex α ∈ NN we have

∂ α (T ∗ S) = (∂ α T ) ∗ S = T ∗ (∂ α S).

Example 47. ∂ α T = T ∗ (∂ α δ0 ).
Theorem 48. (Associativity) Consider R, S, T ∈ D0 (RN ) such that at least two of these
distributions have compact support. Then
R ∗ (S ∗ T ) = (R ∗ S) ∗ T.

Example 49. The above theorem fails if only one of the three distributions has compact
support, even if all convolutions make sense. To see this consider the Heaviside function H .
We have (1∗δ00 )∗H = (10 ∗δ0 )∗H = (0∗δ0 )∗H = 0 and 1∗(δ00 ∗H) = 1∗(δ0 ∗H 0 ) = 1∗(δ0 ∗δ0 ) = 1.

1.5 The Fourier transform

Reference: [1] Chapter 9, [7].
Denition 1. Let f ∈ L1 (RN ). The Fourier transform of f is the function fb : RN −→ RN
dened by Z
fb(ξ) = F(f )(ξ) = e−ix.ξ f (x) dx.
RN

Proposition 2. a) We have kfbkL∞ (RN ) 6 kf kL1 (RN ) .

b) If f ∈ L1 (RN ) then fb is continuous and tends to 0 as |ξ| −→ ∞ .
π N |ξ|2
Example 3. (Gaussian functions) For a > 0 we have F
 2

e−a|x| e−
2
(ξ) = 4a .
a
1
Theorem 4. (Fourier inversion formula) If f ∈ L1 (RN ) and fb ∈ L1 (RN ) then f = F(fb),
Z (2π)N
where Fφ(ξ) = eix.ξ f (x) dx = φ(−ξ).
b
RN
Remark. If f satises the assumptions of the above theorem then necessarily f = g a.e.,
where g is continuous and tends to zero at innity.
Exercise 5. Find the Fourier transform of the following functions of one real variable:
0 if x < 0

1
a) 1[−a,a] ; b) H(x)e −λx
where λ > 0, H(x) = ; c) e−λ|x| ; d) .
1 if x > 1 x2 + λ2

12
 The Schwartz space

Denition 6. S(RN ) := {ϕ ∈ C ∞ (RN ) | xα ∂ β ϕ ∈ L∞ (RN ) for all α, β ∈ NN }.

We dene the following family of seminorms on S(RN ):
X
Np (ϕ) = kxα ∂ β ϕkL∞ (RN ) .
|α|6p, |β|6p

Example 7. C0∞ (RN ) ⊂ S(RN ), Gaussian functions e−a|x|2 (a > 0) belong to S(RN ).
Proposition 8. S(RN ) is stable by dierentiation and multiplication by polynomials. More-
over, for any p ∈ N there exists Cp > 0 such that
X
∀ϕ ∈ S(RN ), kxα ∂ β ϕkL1 (RN ) 6 Cp Np+N +1 (ϕ).
|α|6p, |β|6p

Lemma 9. For any ϕ ∈ S(RN ) the function ϕb is C 1 and ∂j (ϕ)

b = F((−ixj )ϕ).
Lemma 10. For any ϕ ∈ S(RN ) we have F(∂j ϕ) = iξj ϕ.b
Iterating the two lemmas above leads to the following.
Theorem 11. a) For any ϕ ∈ S(RN ) we have ϕb ∈ S(RN ). Moreover, for any p ∈ N there
exists Cp > 0 such that Np (ϕ)
b 6 Cp Np+N +1 (ϕ).
b) F is an isomorphism from S(RN ) to S(RN ) and its inverse isomorphism is 1
(2π)N
F.

Theorem 12. C0∞ (RN ) is dense in S(RN ). More precisely, for any ϕ ∈ S(RN ) there exists a
sequence (ϕk )k>1 ⊂ C0∞ (RN ) such that for any p ∈ N we have Np (ϕk − ϕ) −→ 0 as k −→ ∞.

Temperate distributions

Denition 13. Let u ∈ D0 (RN ). We say that u is a temperate distribution i

∃p ∈ N, ∃C > 0, ∀ϕ ∈ C0∞ (RN ), |hu, ϕi| 6 CNp (ϕ).

Theorem 14. Any temperate distribution u has an extension to a linear form dened on
S(RN ), still denoted u, such that
∀ϕ ∈ S(RN ), |hu, ϕi| 6 CNp (ϕ).

We identify the space of temperate distributions with the space of linear continuous forms on
S(RN ), denoted S 0 (RN ).
Example 15.
a) If f ∈ L1loc (RN ) and there exist C, n such that |f (x)| 6 C(1 + |x|n ), then f ∈ S 0 (RN ).
b) Lp (RN ) ⊂ S 0 (RN ).
c) Any distribution with compact support is a tempered distribution.
Z −ε Z ∞ 
ϕ(x) ϕ(x)
Exercise 16. a) Let u= vp( x1 ) be dened by hu, ϕi = lim dx + dx .
ε↓0 −∞ x ε x
Prove that u ∈ S 0 (R).
b) Prove that ex 6∈ S 0 (R). (Hint: show that for any ϕ ∈ C0∞ (R) and any p ∈ N there exists
C > 0 such that Np (ϕ(· − a)) 6 C(1 + |a|)p .)

13
 ∞
c) Prove that the distribution u = ak δk belongs to S 0 (R) if and only if there exist
X

k=−∞
C > 0, n ∈ N such that |ak | 6 C(1 + |k|n ).
Denition 17. We say that a sequence (uj )j>1
⊂ S 0 (RN ) converges to u ∈ S 0 (RN ) and we
write uj −→ u in i for any ϕ ∈
S 0 (RN ) we have huj , ϕi −→ hu, ϕi.
S(RN )
Remark. uj −→ u in S (R ) implies uj −→ u in D0 (RN ).
0 N

Example 18. a) If fj −→ f in Lp (RN ) then fj −→ f in S 0 (RN ).

b) If fj −→ f a.e. and there exists a polynomial P such that |fj | 6 P a.e. for each j , then
fj −→ f in S 0 (RN ).
Theorem 19. For any u ∈ S 0 (RN ) and any multiindex α we have ∂ α u ∈ S 0 (RN ). Moreover,
if uj −→ u in S 0 (RN ) then ∂ α uj −→ ∂ α u in S 0 (RN ).
Proof.
Denition 20. We say that a function f is slowly increasing as well as its derivatives, and we
denote f ∈ OM (RN ), if f ∈ C ∞ (RN ) and for any multiindex α there exist Cα and mα such
that
|∂ α ϕ(x)| 6 Cα (1 + |x|)mα .

Theorem 21. Let f ∈ OM . Then:

a) For any ϕ ∈ S(RN ) we have f ϕ ∈ S(RN ).
b) For any u ∈ S 0 (RN ) we have f u ∈ S 0 (RN ). Moreover, if uj −→ u in S 0 (RN ) then
f uj −→ f u in S 0 (RN ).
Proof. Exercise. (Hint: Use Leibniz' formula.)

Fourier transform of temperate distributions

Remark 22. If f ∈ L1 (RN ) and ϕ ∈ S(RN ) it is easily seen that hf, ϕi

b = hfb, ϕi.
This inspires the following.
Denition 23. We dene the Fourier transform of a tempered distribution u ∈ S 0 (RN ) by
hb
u, ϕi = hFu, ϕi := hu, ϕi
b for any ϕ ∈ S(RN ).

Theorem 24. We have ub ∈ S 0 (RN ) for any u ∈ S 0 (RN ).

Proof. We have |hu, ϕi|
b 6 Np (ϕ)b 6 C 0 Np+N +1 (ϕ).
Theorem 25. If uj −→ u in S 0 (RN ) then ubj −→ ub in S 0 (RN ).
1
Theorem 26. F is an isomorphism from S 0 (RN ) to S 0 (RN ) and its inverse is F −1 = F.
(2π)N
 N
4π 2 − |ξ|22
 
ε2 |x|2
Example 27. We know that F e − 4
(ξ) = e ε .
ε2
Let ε −→ 0 to get F1 = (2π)N δ0 .
Proposition 28. (Basic properties of the Fourier transform)
a) ub̌ = uˇb where fˇ(x) = f (−x) and hǔ, ϕi = hu, ϕ̌i.
b) F(τa u) = e−ia.ξ Fu.
c) F(eia.x u) = τa ub.
d) F(u( λx )) = |λ|N ub(λξ).

14
Theorem 29. If u is a distribution with compact support then u
b ∈ Om and u
b(ξ) =
hu(x), e−ix.ξ i.
Exercise 30. Prove that
a) F(δ0 ) = 1
b) F(∂ α δ0 ) = i|α| ξ α
c) F(δa ) = e−ia.ξ
d) F(eia.x ) = (2π)N δa
e) F(xα ) = (2π)N i|α| ∂ α δ0
f) Denote by dσR the surface measure of the sphere ∂B(0, R) in R3 .
Show that F(dσR ) = 4πR sin(R|ξ|)
|ξ| .

Theorem 31. The mapping u 7−→ (2π)− 2 Fu is an isometry from L2 (RN ) onto
N
L2 (RN )
and its inverse is (2π) F.
−N
2

Exercise 32. Let u ∈ L2 (RN ). Assume that for a.e. ξ ∈ RN we have

Z
e−ix.ξ u(x) dx −→ g(ξ) as R −→ ∞.
B(0,R)

Show that ub(ξ) = g(ξ) for a.e. ξ.

Theorem 33. Assume that
a) either u, v ∈ L1 (RN ), or
b) u is a distribution with compact support and v ∈ S 0 (RN ).
Then F(u ∗ v) = Fu · Fv.
Corollary 34. For any u ∈ S 0 (RN ) we have
a) F(∂ α u) = i|α| ξ α ub,
b) F(τa u) = e−ia.ξ ub,
c) F(xα u) = i|α| ∂ α ub.
Corollary 35. Assume that u ∈ S 0 (RN ), φ ∈ OM and φb has compact support. Then
F(φu) = (2π)−N φb ∗ u
b.

Exercise 36. Compute F(vp( x1 )).

1.6 Sobolev spaces

Recommended reading: [2] or [3] Chapters VIII and IX.
Denition 1. Let Ω ⊂ RN be an open set and let 1 6 p 6 ∞. The Sobolev space W 1,p (Ω) is
dened by  
∂u
1,p
W (Ω) = u ∈ L (Ω) | p
∈ L (Ω) for i = 1, 2, . . . , N .
p
∂xi
Of course, in the above denition the derivatives are taken in the distributional sense; that
means that there exist g1 , . . . gN ∈ Lp (Ω) such that ∂x
∂u
i
= gi in D0 (Ω), that is
Z Z
∂ϕ
∀ϕ ∈ C0∞ (Ω), u(x) dx = − gi (x)ϕ(x) dx.
Ω ∂xi Ω

The space W 1,p (Ω) is endowed with the norm

N
X ∂u
kukW 1,p = kukLp + .
∂xi Lp
i=1

15
We denote H 1 (Ω) = W 1,2 (Ω). It is a Hilbert space for the scalar product
N D
X ∂u ∂v E
hu, viH 1 = hu, viL2 + , .
∂xi ∂xi L2
i=1

This scalar product induces an equivalent norm on H 1 (Ω).

Proposition 2. a) W 1,p (Ω) is a Banach space for every 1 6 p 6 ∞.
b) W 1,p (Ω) is separable for 1 6 p < ∞.
c) W 1,p (Ω) is reexive for 1 < p < ∞.
d) H 1 (Ω) is a separable Hilbert space.
Denition 3. Assume that Ω is open. We say that an open set ω is compactly included in Ω
and we denote ω ⊂⊂ Ω if ω is compact and ω ⊂ Ω.
Theorem 4. (Friedrichs)
Let u ∈ W 1,p (Ω) with 1 6 p < ∞. There exists a sequence (un ) ⊂ C0∞ (RN ) such that
un |Ω −→ u in Lp (Ω) and
∂un ∂u
|Ω −→ in Lp (ω) for any ω ⊂⊂ Ω and i = 1, . . . , N.
∂xi ∂xi
Idea of the proof. The proof relies on convolution (regularization) and cut-o.
u(x) if x ∈ Ω,

Let ũ(x) = Let (ρn )n>1 be a sequence of molliers. Take ζ ∈ C0∞ (RN )
0 if x ∈ RN \ Ω.
such that ζ = 1 on B(0, 1) and ζ = 0 on RN \ B(0, 2). One can prove that the sequence
x
un (x) = ζ ρn ∗ ũ
n
has the desired properties.
Here we use the
Lemma 5. Let (ρn )n>1 be a sequence of molliers. Let v ∈ Lp (RN ), 1 6 p < ∞. Then
ρn ∗ v ∈ C ∞ (RN ) ∩ Lp (RN ), ρn ∗ v −→ v in Lp (RN ).
Moreover, if v ∈ W 1,p (RN ) then ρn ∗ v ∈ W 1,p (RN ) and ∂
∂xi (ρn ∗ v) = ρn ∗ ∂v
∂xi .

Proposition 6. (dierentiation of a product) Let u, v ∈ W 1,p (Ω) ∩ L∞ (Ω), 1 6 p 6 ∞. Then

uv ∈ W 1,p (Ω) ∩ L∞ (Ω) and

∂ ∂u ∂v
(uv) = v+u , i = 1, . . . , N.
∂xi ∂xi ∂xi

Proposition 7. (dierentiation of a composition) Consider G ∈ C 1 (R) such that G(0) = 0

and G0 is bounded. Then for any u ∈ W 1,p (Ω) with 1 6 p 6 ∞ we have G ◦ u ∈ W 1,p (Ω) and
∂
∂u
(G ◦ u) = G0 ◦ u .
∂xi ∂xi

Proposition 8. (change of variables) Assume that Ω1 , Ω2 ⊂ RN are open sets, H : Ω1 −→ Ω2

is C 1 , bijective, such that H −1 : Ω2 −→ Ω1 is also C 1 and Jac(H) ∈ L∞ (Ω1 ), Jac(H −1 ) ∈
L∞ (Ω2 ), where Jac denotes the Jacobian matrix.

16
 Then for any u ∈ W 1,p (Ω2 ) with 1 6 p 6 ∞ we have u ◦ H ∈ W 1,p (Ω1 ) and
N
∂ X ∂u ∂Hk
(u(H(x))) = (H(x)) (x).
∂xi ∂yk ∂xi
k=1

Idea of the proof. Approximate u by smooth functions, pass to the limit.

Denition 9. Let m ∈ N, m > 2 and 1 6 p 6 ∞. We dene by induction
n o
W m,p (Ω) = u ∈ W m−1,p (Ω) | ∂u
∂xi ∈ W m−1,p (Ω) for all i = 1, . . . , N

u ∈ Lp (Ω) | ∂ α u ∈ Lp (Ω) for any multiindex α ∈ NN with |α| 6 m .


=

The space W m,p (Ω) equipped with the norm

X
kukW m,p = k∂ α ukLp
06|α|6m

is a Banach space.
The space H m (Ω) = W m,2 (Ω) equipped with the scalar product
X
hu, viH m = h∂ α u, ∂ α viL2
06|α|6m

is a Hilbert space.
Theorem 10. (extension) Assume that Ω ⊂ RN is open with ∂Ω bounded and of class C 1 , or
Ω = RN −1 × (0, ∞). Let 1 6 p 6 ∞. There exists a linear operator P : W 1,p (Ω) −→ W 1,p (RN )
such that for any u ∈ W 1,p (Ω),
P u|Ω = u,
kP ukLp (RN ) 6 CkukLp (Ω) ,
kP ukW 1,p (RN ) 6 CkukW 1,p (Ω) .

Corollary 11. Assume that Ω is open and ∂Ω is of class C 1 . Let 1 6 p < ∞. There exists a
sequence (un )n>1 ⊂ C0∞ (RN ) such that un |Ω −→ u in W 1,p (Ω).

Sobolev inequalities

Theorem 12. (Sobolev, Gagliardo, Nirenberg) Let 1 6 p < N. Then W 1,p (RN ) ⊂ Lp∗ (RN ),
1 1 1
where p∗ is given by = − . Moreover, there exists a constant C = C(p, N ) such that
p∗ p N

kukLp∗ (RN ) 6 Ck∇ukLp (RN ) .

Remark 13. The value of p∗ is easily obtained by scaling.

∂u
Remark 14. A stronger inequality is true: whenever u ∈ L1loc (RN ) satises ∈ Lp (RN )
∂xi
for all 1 6 i 6 N , there exists a constant u∞ such that u − u∞ ∈ Lp (RN ) and
∗

ku − u∞ kLp∗ (RN ) 6 Ck∇ukLp (RN ) .

17
Corollary 15. Let 1 6 p < N. Then W 1,p (RN ) ⊂ Lq (RN ) for any q ∈ [p, p∗ ] and the
embedding is continuous.
Theorem 16. (the case p = N ). We have W 1,N (RN ) ⊂ Lq (RN ) for any q ∈ [n, ∞) and the
embedding is continuous.
Theorem 17. (Morrey) If p > N we have W 1,p (RN ) ⊂ L∞ (RN ) and the embedding is
continuous.
Moreover, for any u ∈ W 1,p (RN ) there exists a continuous function ũ such that u = ũ
almost everywhere on RN and
|ũ(x) − ũ(y)| 6 C|x − y|α k∇ukLp for all x, y ∈ RN ,
where α = 1 − Np and C is a constant depending only on N and p.
Corollary 18. We get Sobolev inequalities for higher order Sobolev spaces by repeatedly
applying the above inequalities. For instance,
1 m 1 1 m
• if − > 0 and = − then W m,p (RN ) ⊂ Lq (RN ),
p N q p N
1 m
• if − = 0 then W m,p (RN ) ⊂ Lq (RN ) for all p 6 q < ∞,
p N
1 m
• if − < 0 then W m,p (RN ) ⊂ L∞ (RN )
p N
and all these embeddings are continuous. h i
Moreover, if m − Np > 0 is not an integer and k = m − Np we have W m,p (RN ) ⊂ C k (RN ).
Corollary 19. Assume that Ω ⊂ RN is open with ∂Ω bounded and of class C 1 , or Ω =
RN −1 × (0, ∞). Let 1 6 p 6 ∞. We have:

where if 1 6 p < N,
∗ 1 1 1
W 1,p (Ω) ⊂ Lp (Ω) p∗ = p − N

W 1,N (Ω) ⊂ Lq (Ω) for all q ∈ [N, ∞) in the case p = N,

W 1,p (Ω) ⊂ L∞ (Ω) if p > N

and all these embeddings are continuous.
Furthermore, if p > N there exists a continuous function ũ such that u = ũ a.e. and
∀x, y ∈ Ω, |ũ(x) − ũ(y)| 6 C|x − y|α kukW 1,p

where α = 1 − Np and C is a constant depending only on Ω, N and p.

Corollary 20. If Ω ⊂ RN is an open set with suciently smooth boundary, we have inequal-
ities for the Sobolev spaces W m,p (Ω) similar to those for W m,p (RN ).
Theorem 21. (Rellich-Kondrachov). Assume that Ω ⊂ RN is bounded with C 1 boundary.
The following embeddings are compact:
W 1,p (Ω) ⊂ Lq (Ω) for all q ∈ [1, p∗ [ where 1
p∗ = 1
p − 1
N if 1 6 p < N,

W 1,N (Ω) ⊂ Lq (Ω) for all q ∈ [N, ∞) in the case p = N,

W 1,p (Ω) ⊂ C(Ω) if p > N.

18
 The space W01,p (Ω)

Denition 22. Let 1 6 p < ∞. The space W01,p (Ω) is the closure of C0∞ (Ω) in W01,p (Ω).
The space H01 (Ω) is the closure of C0∞ (Ω) in H 1 (Ω).
Remark. We know that W01,p (RN ) = W 1,p (RN ). In general W01,p (Ω) W 1,p (Ω)!
Lemma 23. Assume that 1 6 p < ∞, u ∈ W 1,p (Ω) and supp(u) is a compact subset of Ω.
Then u ∈ W01,p (Ω).
Theorem 24. Suppose that 1 6 p < ∞, ∂Ω is of class C 1 and u ∈ W 1,p (Ω) ∩ C(Ω). Then
u ∈ W01,p (Ω) ⇔ u = 0 on ∂Ω.

Theorem 25. Assume that ∂Ω is of class C 1 and 1 < p < ∞. The following properties are
equivalent:
(i) u ∈ W01,p (Ω);
(ii) there exists a constant C such that
Z
∂ϕ 1 1
u dx 6 CkϕkLp0 (Ω) for all ϕ ∈ C0∞ (RN ) and i = 1, . . . , N, where + 0 = 1.
Ω ∂xi p p

u(x) if x ∈ Ω,
 ^
(iii) The function ũ(x) = belongs to W 1,p (RN ) and ∂
∂xi (ũ) = ∂u
.
0 if x ∈ RN \ Ω ∂xi

Theorem 26. (Poincaré inequality). Assume that Ω ⊂ RN is bounded with C 1 boundary

and 1 < p < ∞. There exists a constant C depending only on Ω and p such that
kukLp (Ω) 6 Ck∇ukLp (Ω) for all u ∈ W01,p (Ω).

In particular, k∇ukLp (Ω) is a norm on W01,p (Ω) equivalent to kukW 1,p .

Proof. We will prove the following stronger result:
Lemma 27. Under the assumptions of the above theorem, let V be a closed vector subspace
of W 1,p (Ω) such that 1Ω (the constant function equal to 1 on Ω) does not belong to V . Then
there exists C > 0 such that for any v ∈ V there holds
kvkLp (Ω) 6 Ck∇vkLp (Ω) .

Proof. We argue by contradiction and assume that the lemma is false. Then for each
n ∈ N∗ there exists vn ∈ V such that kvn kLp = 1 and k∇vn kLp 6 n1 . The sequence (vn )n>1
is bounded in W 1,p (Ω). Since W 1,p (Ω) is reexive and separable, there exists a subsequence
(vnk )k>1 that converges weakly to some v ∈ W 1,p (Ω). We have vnk ∈ V for all k and V is
a closed convex subset in W 1,p (Ω), hence it is weakly closed; thus v ∈ V . By the Rellich-
Kondrachov theorem the inclusion W 1,p (Ω) ⊂ Lp (Ω) is compact, hence vnk −→ v strongly in
Lp (Ω) and we infer that kvkLp = limk→∞ kvnk kLp = 1. On the other hand we have ∇vnk * ∇v
weakly in Lp (Ω), hence
k∇vkLp 6 lim inf k∇vnk kLp = 0.
k→∞

Thus ∇v = 0 a.e. and this implies that there exists a constant k such that v = k a.e. Since
kvkLp = 1 we have k 6= 0 and consequently 1Ω = k1 v ∈ V , a contradiction. 

19
Corollary 28. (second Poincaré inequality) Assume that Ω ⊂ RN isZbounded with C 1
1
boundary and 1 < p < ∞. For any f ∈ L1 (Ω) we denote by m(f ) = f (x) dx the mean
|Ω| Ω
value of f on Ω. There exists a constant C (depending only on Ω and p) such that
Z Z
p
|u(x) − m(u)| dx 6 C |∇u(x)|p dx.
Ω Ω

Proof. Exercise. (Hint: Use the Lemma for the subspace V = {u ∈ W 1,p (Ω) | m(u) = 0}. )
Remark 29. Poincare's inequality is still true if Ω has nite measure or if Ω has bounded
projection on some axis. It is also true for p = 1 (with a dierent proof).
Denition 30. The space W −1,p0 (Ω) is the dual space of W01,p (Ω), 1 6 p < ∞.
Proposition 31. Let F ∈ W −1,p0 (Ω). There exist functions f0 , f1 , . . . , fN ∈ Lp0 (Ω) such that
Z N Z
∂v
for all v ∈ W01,p (Ω).
X
hF, vi = f0 v dx + fi dx
Ω Ω ∂xi
i=1

Moreover, if Ω is bounded we can take f0 = 0.

1.7 Exercises
Exercise 1. (simple layer of density f ) Let f ∈ L1loc (R2 ). For any ϕ ∈ C0∞ (R3 ) dene
Z
hµ, ϕi = f (x, y)ϕ(x, y, 0) dx dy.
R2

a) Show that µ ∈ D0 (R3 ) and µ is of order 0. The distribution µ is called simple layer of
density f supported in the plane {z = 0}.
b) Let un be the distribution consisting in a simple layer of density nf supported in the
plane {z = n1 } plus a simple layer of density −nf supported in the plane {z = 0}. Prove that
un −→ u, where Z
∂ϕ
hu, ϕi = f (x, y) (x, y, 0) dx dy.
R2 ∂z
Show that u is a distribution of order 1 (called double layer).
c) Prove that u = − ∂µ∂z . Assuming in addition that f is C , compute
1 ∂µ
∂x and ∂µ
∂y and show
that these are simple layer distributions.
Exercise 2. A function f : (a, b) −→ R is piecewise C m i there exists a nite number of
points a = a0 < a1 < a2 < · · · < an = b such that f is C m on each of the intervals (ai−1 , ai ),
and f as well as all its derivatives up to order m admit continuous extensions to the intervals
(a0 , a1 ], [ai , ai+1 ], [an−1 , an ). For 0 6 k 6 m we denote by f (k) (ak ±) the limit to the right
(resp. to the left) of f (k) at ak .
a) Assume that f is piecewice C 1 in (a, b). Prove that, with the above notation, we have
N
X −1
f 0 = {f 0 } + (f (ai + 0) − f (ai − 0)) δai ,
i=1

where f 0 is the derivative of f in D0 ((a, b)) and {f 0 } is the "usual" derivative of f in (a, b) \
{a1 , . . . , an−1 }.

20
 b) Assume that f is piecewise C 2 in (a, b). Prove that
N
X −1 N
X −1
00 00 0 0
(f (ai + 0) − f (ai − 0)) δa0 i .


f = {f } + f (ai + 0) − f (ai − 0) δai +
i=1 i=1

Exercise 3. Suppose that Ω ⊂ RN is open, bounded, with C 1 boundary. Let Ωc = RN \ Ω.

Suppose that f : RN −→ R is C 1 on Ω and on Ωc and its restrictions f|Ω and f|Ωc admit
continuous extensions to functions in C 1 (Ω) and C 1 (Ωc ), respectively. We denote by fint and
fext , respectively, these extensions on ∂Ω. Prove that
 
∂f ∂f
= + [fext − fint ]ni dσ,
∂xi ∂xi

where { ∂x
∂f
i
} is the usual derivative of f in Ω ∪ Ωc , and [fext − fint ]ni dσ is the simple layer
distribution supported by ∂Ω with density [fext − fint ]ni = [fext − fint ]hn, ei i.
N
∂2
Exercise 4. Assume that u ∈ satises ∆u = 0 in S 0 (RN ), where ∆ =
X
S 0 (RN ) .
i=1
∂x2i
a) Prove that |ξ|2 ub = 0 in RN .
b) Show that supp(b u) ⊂ {0}.
c) Show that there exist k ∈ N and aα ∈ C such that ub = |α|6k aα ∂ α δ0 .
P
d) Prove that there exists a polynomial P satisfying ∆P = 0 such that u = P.
Exercise 5. Consider the Poisson equation −∆u = f in S 0 (RN ), where N > 3. Assume that
f ∈ S 0 (RN ) and fb is a bounded function in a neighborhood of the origin.
1
a) Show that 2 fb is a tempered distribution.
|ξ|
b) Prove that ub = |ξ|12 fb + u0 in S 0 (RN ), where supp(u0 ) ⊂ {0}.
 
c) Deduce that u = F −1 |ξ|12 fb + P , where P is a harmonic polynomial (i.e. ∆P = 0).
Exercise 6. Prove that for any f ∈ S 0 (RN ) and any λ > 0 the equation −∆u + λu = f has a
unique solution in S 0 (RN ).
√
Exercise 7. For ε > 0, let Fε (t) = t2 + ε2 − ε, so that Fε ∈ C 1 (R) and F (0) = 0. Let
1 ≤ p ≤ ∞ and let u ∈ W 1,p (Ω) be a real-valued function.
a) Prove that for any ε > 0 we have Fε ◦ u ∈ W 1,p (Ω) and compute ∂
∂xj (Fε ◦ u).

b) Consider a sequence εn −→ 0 and use part (a) to show that |u| ∈ W 1,p (Ω) and ∂|u|
∂xj =
sgn(u) ∂x
∂u
j
almost everywhere. Deduce that u+ ∈ W 1,p (Ω) and ∂u
∂xj = 1{u>0} ∂xj a.e.
+ ∂u

Exercise 8. (trace operator) Assume that 1 6 p < ∞. For x = (x1 , . . . , xN −1 , xN ) ∈ RN , we

denote x0 = (x1 , . . . , xN −1 ) ∈ RN −1 . We denote RN
+ =R
N −1 ×]0, ∞[.

a) Prove that there exists C > 0 such that for any u ∈ C01 (RN ) there holds
Z 1
p
0 p 0
|u(x , 0)| dx ≤ CkukW 1,p (RN ) .
+
RN −1

Hint. Let G(t) = |t|p−1 t. Then G ∈ C 1 (R) and for any u ∈ C)1 (RN ) we have
Z ∞ Z ∞
0 ∂ ∂u 0
G(u(x0 , xn )) dxN = − G0 (u(x0 , xn ))


G(u(x , 0)) = − (x , xN ) dxN .
0 ∂xN 0 ∂xN

21
Use Hölder's inequality to show that there exists C > 0 such that for any x0 ∈ RN −1 ,
Z ∞
0 ∂u 0 p
p
|u(x , 0)| ≤ C |u(x0 , xn )|p + (x , xN ) dxN .
0 ∂xN

Integrate over RN −1 .
b) Prove that the mapping u 7−→ u(·, 0) can be extended to a continuous linear mapping
Tr : W 1,p (RN p
+ ) −→ L (R
N −1 ) and that W 1 (RN ) ⊂ Ker(Tr).
0 +

c) From now on we assume that 1 < p < ∞. Let p0 = p−1 p

be the conjugate exponent of p.
Prove that for any u ∈ W (R ) and any v ∈ W (R ) there holds
1,p N 1,p 0 N

Z Z Z
∂u ∂v
v dx = − u dx − Tr(u)(x0 )Tr(v)(x0 )dx0
RN
+
∂xN RN
+
∂xN RN −1

Z Z
∂v
= − u dx + Tr(u)(x0 )Tr(v)~n.e~N dσ,
RN
+
∂xN ∂RN
+

where ~n = (0, −1) is the outer normal to RN+ at any point (x , 0) ∈ ∂R+ and σ is the surface
0 N

measure on ∂R+ , that is the Lebesgue measure on R .

N N −1

Hint. Prove the above formula for u, v ∈ C01 (RN ), then use density theorems.
u(x) if xN > 0,

d) Let u ∈ W 1,p (RN ) such that Tr(u) = 0. Let ũ(x) = Prove that
0 if xN ≤ 0.
+ ), and then conclude that Ker(Tr) = W0 (R ).
ũ ∈ W 1,p (RN ). Show that u ∈ W01,p (RN 1,p N

Hint. Use c) and the theorem on the characterization of W0 functions.

1,p

22
2 Second order elliptic equations

2.1 Laplace's equation and Poisson's equation

References for this section: [4] Section 2.2. See also [5] Chapter 8.
Laplace's equation: ∆u = 0.
Poison's equation: ∆u = f.
Physical interpretation.
Denition 1. A C 2 function satisfying the Laplace equation is called a harmonic function.
Laplace's equation is invariant under rotations, and then it is natural to seek for radial
solutions u(x) = v(r) where r = |x|. A simple computation gives ∆u(x) = v 00 (r) + N r−1 v 0 (r)
and we infer that all radial solutions of Laplace's equation in RN \ {0} are of the form v(r) =
A log r + B if N = 2, respectively v(r) = rNA−2 + B if N > 3.
Denition 2. The function Φ : RN \ {0} −→ R,
1


 − ln |x|
 2π
 if N = 2,
Φ(x) =
1 1
if N > 3,



N (N − 2)ωN |x|N −2


where ωN = |BRN (0, 1)| is the volume of the unit ball in RN , is the fundamental solution of
Laplace's equation.
It is clear that Φ ∈ C ∞ (RN \ {0}) and ∆Φ = 0 in RN \ {0}.
Exercise 3. Prove that Φ ∈ L1loc (RN ), then show that Φ denes a tempered distribution and
that ∆Φ = δ0 in S 0 (RN ).
Theorem 4. Assume that f ∈ C02 (RN ). Let u = Φ ∗ f , that is
Z Z
(1) u(x) = Φ(x − y)f (y) dy = Φ(y)f (x − y) dy.
RN RN

Then u ∈ C 2 (RN ) and ∆u = f in RN . In other words u solves Poisson's equation in RN .

Proof. Exercise. Proceed as follows.
Step 1. Prove that u given by (1) is C 2 and ∂u
∂xj =Φ∗ ∂xj , ∂xk ∂xj
∂f ∂2u
=Φ∗ ∂xk ∂xj .
∂2f

Step 2. From step 1 we have

Z Z
∆u(x) = Φ(y)∆x f (x − y) dy + Φ(y)∆x f (x − y) dy.
B(0,ε) RN \B(0,ε)

Show that the rst integral tends to 0 as ε −→ 0.

Step 3. We have ∆x f (x − y) = ∆y f (x − y). Using integration by parts,
Z
Φ(y)∆x f (x − y) dy
RN \B(0,ε)
Z Z
=− ∇Φ(y).∇y f (x − y) dy + Φ(y)(∇y f (x − y).~n(y)) dσ(y),
RN \B(0,ε) ∂(RN \B(0,ε))

where ~n(y) = − |y|

y
is the outward normal to ∂(RN \ B(0, ε)) at y ∈ S(0, ε). Show that the
surface integral in the above formula tends to 0 as ε −→ 0.

23
 Step 4. Perform another integration by parts to discover
Z
− ∇Φ(y).∇y f (x − y) dy
RN \B(0,ε)
Z Z
= ∆Φ(y)f (x − y) dy − (∇Φ(y).~n(y))f (x − y) dσ(y).
RN \B(0,ε) ∂(RN \B(0,ε))

The rst integral is 0 because

 ∇Φ = 0 on R \ {0}. We have Φ(x) = φ(|x|) = φ(r) and then
N

 − 2π|y| if N = 2,
1

∇Φ(y).~n(y) = −ϕ0 (|y|) = for any y ∈ S(0, ε). The surface measure
 −
 1
N ωN |y|N −1
if N ≥ 3
of ∂(RN \ B(0, ε)) is precisely N ωN εN −1 . Using the continuity of f , the surface integral in the
last formula tends to f (x) as ε −→ 0.

2.2 Mean-value formulae

Theorem 5. Let Ω ⊂ RN be an open set. Assume that u ∈ C 2 (Ω) is a harmonic function in
Ω. Then for any ball B(x, r) ⊂ Ω we have
Z Z
1 1
u(x) = u(y) dσ(y) = u(y) dy.
|∂B(x, r)| ∂B(x,r) |B(x, r)| B(x,r)

Proof.
Theorem 6. (converse to the mean-value theorem) a) If u ∈ C 2 (Ω) satises
Z
1
u(x) = u(y) dσ(y) for any ball B(x, r) ⊂ Ω
|∂B(x, r)| ∂B(x,r)

then u is harmonic in Ω.
b) If u ∈ C 2 (Ω) satises
Z
1
u(x) = u(y) dy for any ball B(x, r) ⊂ Ω
|B(x, r)| B(x,r)

then u is harmonic in Ω.
Proof.
Theorem 7. (strong maximum principle) Assume that Ω is connected and u ∈ C 2 (Ω) is
harmonic in Ω. If there exists x0 ∈ Ω such that u(x0 ) = max u then u is constant in Ω.
Ω
Moreover, if Ω is bounded and u ∈ C 2 (Ω) ∩ C(Ω) is harmonic in Ω then
max u = max u.
Ω ∂Ω

Proof.
Remark 8. A similar statement holds with min instead of max .
Corollary 9. Let Ω ⊂ RN be an open, bounded, connected set. Assume that u ∈ C 2 (Ω)∩C(Ω)
satises
∆u = 0 in Ω


u=g on ∂Ω,
where g > 0 on ∂Ω and g 6= 0. Then u(x) > 0 for any x ∈ Ω.

24
Corollary 10. Assume that Ω ⊂ RN is open, connected and bounded. Let g ∈ C(∂Ω) and
f ∈ C(Ω). There exists at most one solution u ∈ C 2 (Ω) ∩ C(Ω) to the boundary value problem
−∆u = f in Ω


u=g on ∂Ω.

Theorem 11. (regularity)

Z If Ω ⊂ RN is open and u ∈ C(Ω) satises the mean value property
1
u(x) = u(y) dσ(y) for each ball B(x, r) ⊂ Ω, then u ∈ C ∞ (Ω).
|∂B(x, r)| ∂B(x,r)
Proof.
Warning. The function u does not necessarily admit a continuous extension to ∂Ω.
Theorem 12. (estimates on derivatives) Assume that Ω ⊂ RN is open and u is harmonic in
Ω. Then for each k ∈ N there exist constants Ck depending only on N and k such that for any
ball B(x0 , r) ⊂ Ω and for any multiindex α with |α| = k we have
Ck
|∂ α u(x0 )| 6 kukL1 (B(x0 ,r)) .
rN +k
N +1 k
(We may take C0 = ω1N and Ck = (kN 2ωN ) for k > 1.)
Proof.
Theorem 13. (Liouville) If u : RN −→ R is harmonic and bounded, then u is constant.
Proof.
Corollary 14. Let N > 3 and f ∈ C02 (RN ). Then any bounded solution of Poisson's
equation −∆u = f in RN is of the form
u=Φ∗f +C

for some constant C .

Proof.
Theorem 15. (Harnack's inequality) For each connected open set ω ⊂⊂ Ω there exists
C > 0, depending only on ω and Ω, such that for any nonnegative harmonic function u in Ω
we have
sup u 6 C inf u.
ω ω

Proof.

2.3 Green's function

Our purpose is to obtain representation formulae for solutions of the boundary-value problem
in Ω,

−∆u = f
(2)
u=g on ∂Ω.
We will use Green's formula: if Ω is bounded with C 1 boundary and u, v ∈ C 2 (Ω) ∩ (Ω) there
holds Z Z
∂v ∂u
u∆v − v∆u dx = u −v dσ.
Ω ∂Ω ∂n ∂n
Let u ∈ Fix x ∈ Ω, ε > 0 such that B(x, ε) ⊂ Ω. Integrating by parts on Ωε :=
C 2 (Ω).
Ω \ B(x, ε) we get
Z Z
∂Φ ∂u
u(y)∆Φ(y − x) − Φ(y − x)∆u(y) dy = u(y) (y − x) − Φ(y − x) (y) dσ(y),
Ωε ∂Ωε ∂n ∂n

25
where n denotes the unit normal vector on ∂Ωε and σ is the surface measure. Recall that
∆Φ = 0 in RN \ {0}. Passing to the limit in the above identity and using standard estimates
we get, for any x ∈ Ω and any u ∈ C 2 (Ω),
Z Z
∂Φ ∂u
u(x) = u(y) (y − x) − Φ(y − x) (y) dσ(y) − Φ(y − x)∆u(y) dy.
∂Ω ∂n ∂n Ω

∂u
Problem: We only know that u solves (2), but we do not know on ∂Ω.
∂n
Idea: introduce a corrector function φx = φx (y) such that
∆φx = 0 in Ω,


φx (y) = Φ(y − x) on ∂Ω.

Then
∂φx
Z Z
x ∂u
− φ (y)∆u(y) dy = u(y) (y) − Φ(y − x) (y) dσ(y).
Ω ∂Ω ∂n ∂n

Denition 1. Green's function for the domain Ω is G : Ω × Ω \ {(x, x) | x ∈ Ω} −→ R,

G(x, y) = Φ(y − x) − φx (y).

in D0 (Ω),

−∆y G(x, ·) = δx
Exercise 2. Fix x ∈ Ω. Prove that G(x, y) = 0 for y ∈ ∂Ω.
Theorem 3. (Representation formula using Green's function) If u ∈ C 2 (Ω) solves (2) then
Z Z
∂G
u(x) = − g(y) (x, y) dσ(y) + f (y)G(x, y) dy.
∂Ω ∂ny Ω

Green's function for the half-space RN+ = {x = (x1 , . . . , xN ) ∈ RN | xN > 0}.

For x = (x1 , . . . , xN ) ∈ we denote by x̃ = (x1 , . . . , xN −1 , −xN ) the point obtained by
RN
reecting x with respect to the hyperplane {xN = 0}.
Fix x ∈ RN+ . The function y 7−→ Φ(y − x̃) is harmonic in R+ and Φ(y − x̃) = Φ(y − x) if
N

y ∈ ∂R+ (or equivalently if yN = 0).

N

Denition 4. Green's function for the half-space RN+ is G(x, y) = Φ(y − x) − Φ(y − x̃).
∂G ∂G 2xN 1
A straightforward computation gives (x, y) = − (x, y) = − .
∂ny ∂yN N ωN |x − y|N
Denition 5. The function K : RN+ × ∂RN+ −→ R,

2xN 1
K(x, y) =
N ωN |x − y|N

is the Poisson kernel for RN

+.

Theorem 6. (Poisson's formula for a half-space) Assume that g is a continuous and bounded
function on RN −1 = ∂RN
+ . For any x ∈ R+ dene
N

g(y 0 )
Z Z
2xN g(y) 2xN
u(x) = dσ(y) = dy 0 .
N ωN ∂RN
+
|x − y|N N ωN RN −1 |x − (y 0 , 0)|N

Then
(i) u ∈ C ∞ (RN ∞ N
+ ) ∩ L (R+ ),

26
 (ii) ∆u = 0 in RN
+,

(iii) lim u(x) = g(x∗ ) for any x∗ ∈ ∂RN

+.
x→x∗ , x∈RN
+

in RN

+, ∆u = 0
In other words, u solves the boundary-value problem
on ∂RN +. u=g
Proof. Direct computation using dierentiation under integral sign and integration by
parts. See [4] section 2.2.4 (b).

2.4 More general second-order linear elliptic equations

References: [4] Chapter 6, [3] sections 9.5 to 9.8, [5] Chapter 8.
Theorem 1. (Dirichlet's principle) Let Ω ⊂ RN be Za bounded domain with C 1 boundary.
1
Denote A = {v ∈ C 2 (Ω) | v = g on ∂Ω}. Let I(v) = |∇v|2 (x) − f (x)v(x) dx.
2 Ω
Then u ∈ C 2 (Ω) solves the boundary-value problem (2) if and only if u minimizes the
functional I in the set A.
Proof. "⇒" Let w ∈ A. Then
Z
(−∆u(x) − f (x))(u(x) − w(x)) dx = 0.
Ω

Integrating by parts we get

Z XN
∂u ∂
· (u − w) − f (u − w) dx = 0,
Ω ∂xi ∂xi
i=1

that is
Z Z Z
2 1 1
|∇u| − f u dx = ∇u.∇w − f w dx 6 |∇u|2 + |∇w|2 − f w dx,
Ω Ω Ω 2 2
and consequently I(u) 6 I(w).
Exercise 2. Prove that there is at most one solution u ∈ C 2 (Ω) to the boundary-value problem
(2).
(Hint: if u and v are two solutions, let w = u − v . Show that ∆w = 0 in Ω. Multiply this
equality by w and integrate by parts in Ω. )
In this section we consider second-order partial dierential operators
N   N
∂u ∂u
(3) (divergence form),
X X
Lu = − ai,j (x) + bi (x) + c(x)u
∂xi xj ∂xi
i,j=1 i=1

N N
∂2u ∂u
(4) (nondivergence form).
X X
Lu = − ai,j (x) + bi (x) + c(x)u
∂xi ∂xj ∂xi
i,j=1 i=1

Exercise 3. Assume that ai,j are C 1 functions. Prove that after replacing the coecients bi (x)
by some other functions b̃i (x), an operator in divergence form can be written into nondivergence
form and vice-versa.
We assume the symmetry condition ai,j = aj,i (this is always possible after replacing ai,j
by 2 (ai,j + aj,i )).
1

27
Denition 4. The partial dierential operator L is uniformly elliptic in Ω if there exists α > 0
such that
N
X
∀x ∈ Ω, ∀ξ ∈ RN , ai,j (x)ξi ξj > α|ξ|2 .
i,j=1

Denition 5. The bilinear form associated with the divergence-form second order partial
dierential operator L in (3) is
N
Z X N
∂u ∂v X ∂u
B(u, v) = ai,j (x) + bi (x) v + c(x)uv dx.
Ω i,j=1 ∂xi ∂xj ∂xi
i=1

Let L be an operator in divergence form as above. Consider the boundary value problem
Lu = f in Ω

(5)
u=0 on ∂Ω.
Denition 6. We say that u ∈ H01 (Ω) is a weak solution of (5) if
B(u, v) = hf, vi
for all v ∈ H01 (Ω).
Exercise 7. Assume that u ∈ C 2 (Ω) ∩ C(Ω) is a solution of (5). Prove that u is a weak
solution of (5).
Theorem 8. (Lax-Milgram) Let H be a Hilbert space. Assume that B : H × H −→ R is a
bilinear form and there exist C, β > 0 such that
∀u, v ∈ H, |B(u, v)| 6 Ckuk · kvk (B is continuous)
and
∀u ∈ H, |B(u, u)| > βkuk2 , (B is coercive).
Let f : H −→ R be a bounded linear functional on H .
Then there exists a unique element u ∈ H such that B(u, v) = hf, vi for all v ∈ H .
Remark. If B is symmetric, this is simply Riesz' Theorem. Lax-Milgram Theorem does
not require the symmetry of B .
Theorem 9. Assume that Ω is bounded with C 1 boundary, L is uniformly elliptic in Ω and
ai,j , bi , c ∈ L∞ (Ω). There exist C, β, γ > 0 such that for all u, v ∈ H01 (Ω) we have
|B(u, v)| 6 CkukH01 · kvkH01 and
βkuk2H 1 6 B(u, u) + γkuk2L2 .
0

Proof.
Theorem 10. Under the assumptions of the previous theorem, there exists γ > 0 such that
for any µ > γ and for any f ∈ H −1 (Ω) there exists a unique weak solution u ∈ H01 (Ω) of the
boundary-value problem
Lu + µu = f in Ω


u = 0 on ∂Ω.
Proof.
For µ > γ the mapping Lµ = L + µId is an isomorphism from H01 (Ω) to H −1 (Ω).
If bi = 0 for all i and c > 0 the above results hold with γ = 0.
Question. Are weak solutions smooth?
Example: −∆u = f in RN . If f ∈ H k (RN ) we may control the L2 −norm of derivatives of
order k + 2 on u in terms of k∇k ukL2 (this can be easily seen by using the Fourier transform).

28
2.5 Exercises
Exercise 1. (symmetry of Green's function). Let Ω ⊂ RN be a domain and let G be its Green
function. Our aim is to show that
G(x, y) = G(y, x) for all x, y ∈ Ω, x 6= y.
a) Let v(z) = G(x, z) and w(z) = G(y, z). Fix ε > 0 such that B(x, ε) ⊂ Ω and B(y, ε) ⊂ Ω.
Show that
Z Z
∂v ∂w ∂w ∂v
(z) · w(z) − (z) · v(z) dσ(z) = (z) · v(z) − (z) · w(z) dσ(z).
∂B(x,ε) ∂n ∂n ∂B(y,ε) ∂n ∂n

(Hint. Use Green's identity on Ω \ (B(x, ε) ∪ B(y, ε)) and recall that ∆v = ∆w = 0.)
b) Pass to the limit as ε −→ 0 to get the desired conclusion.
Exercise 2. (Green's function for a ball)
a) Given x ∈ RN \ {0}, the point x∗ = |x|x2 is called the point dual to x with respect to the
unit sphere ∂B(0, 1). Check that for any x ∈ RN \ {0} and any y ∈ RN with |y| = 1 there
holds |x| · |y − x∗ | = |y − x|. (Hint: compute the square of the left-hand side.)
b) Let x ∈ B(0, 1), x 6= 0. Let φx (y) = Φ (|x|(y − x∗ )) . Show that φx is harmonic in B(0, 1)
and φx (y) = Φ(x − y) for all y ∈ ∂B(0, 1). Deduce that the Green function for the unit ball is
G(x, y) = Φ(y − x) − Φ (|x|(y − x∗ )) , where x, y ∈ B(0, 1), x 6= y, x 6= 0.
c) Show that for any y ∈ ∂B(0, 1) we have
N
∂G X ∂G −1 1 − |x|2
(x, y) = yi (x, y) = .
∂ny ∂yi N ωN |x − y|N
i=1

d) Assume that u ∈ C 2 (B(0, 1)) solves the boundary-value problem

in B(0, 1),

∆u = 0
(6)
u=g on ∂B(0, 1).
Prove the representation formula
1 − |x|2
Z
g(y)
u(x) = dσ(y).
N ωN ∂B(0,1) |x − y|N

e) Let R > 0 and suppose that u ∈ C 2 (B(0, R)) solves the boundary-value problem
in B(0, R),

∆u = 0
u=g on ∂B(0, R).
Prove that ũ(x) = u(Rx) solves (6) with g̃(x) = g(Rx) instead of g . Then change variables to
obtain Poisson's formula for the ball B(0, R) :
R2 − |x|2
Z
g(y)
(7) u(x) = dσ(y) for all x ∈ B(0, R).
N ωN ∂B(0,R) |x − y|N

f) Assume that g ∈ C(∂B(0, R)) and dene u by (7). Prove that

(i) u ∈ C ∞ (B(0, R)),
(ii) ∆u = 0 in B(0, R),

29
 (iii) lim u(x) = g(x0 ) for any x0 ∈ ∂B(0, R).
x→x0 , x∈B(0,R)

Exercise 3. Consider the nonhomogeneous Dirichlet boundary-value problem

in Ω

Lu = f
(8)
u=g on ∂Ω.
where L is an elliptic operator in divergence form as in (3), f ∈ H −1 (Ω) and there exists u0 ∈
H 1 (Ω) such that u0 = g on ∂Ω. Assume that L is uniformly elliptic in Ω and ai,j , bi , c ∈ L∞ (Ω).
a) Prove that Lu0 ∈ H −1 (Ω).
Lv = f − Lu0 in Ω

b) Prove that u solves (8) i v := u − u0 solves
v=0 on ∂Ω.
c) Infer that whenever we are able to solve (5) for all f ∈ H −1 (Ω), we are also able to solve
(8) for all f ∈ H −1 (Ω).
Exercise 4. Let Ω ⊂ RN be a bounded domain with C 1 boundary. Consider the nonhomoge-
neous Neumann boundary-value problem
in Ω

−∆u = f
(9) ∂u
∂n = g on ∂Ω.
a) Assume that (9) admits a solution u ∈ C 2 (Ω). Show that f and g must satisfy
Z Z
(10) f (x) dx + g(x)dσ(x) = 0.
Ω ∂Ω

and that u + constant is also a solution. Then prove that for any v ∈ C 1 (Ω) we have
Z XN Z Z
∂u ∂v
(11) dx = f v dx + gv dσ.
Ω ∂xi ∂xi Ω ∂Ω
i=1

Show that (11) holds for any v ∈ H 1 (Ω). (Hint: use the density of smooth functions in H 1 (Ω)
and the fact that the mapping ϕ 7−→ ϕ|∂Ω extends to a continuous mapping from H 1 (Ω) to
L2 (Ω) - see LemmaIX.9 in [2] 1 ). Z 
Let Hm
1 (Ω) := ϕ ∈ H 1 (Ω) 1 (Ω) is a
ϕ(x) dx = 0 . We say that a function u ∈ Hm
Ω
weak solution of (9) i (11) holds for any v ∈ Hm
1 (Ω).

b) Assume that f ∈ L2 (Ω), g is continuous on ∂Ω, (10) is satised and u ∈ C 1 (Ω) is a weak
∂u
solution of (9). Prove that −∆u = f a.e. in Ω and = g on ∂Ω. (Hint: Show that (11)
∂n
holds
Z for any vZ∈ H 1 (Ω). Take v ∈ C0∞ (Ω) and deduce −∆u = f a.e. in Ω. Then show that
∂u ∂u
ϕ dσ = gϕ dσ for any ϕ ∈ C ∞ (RN ) and deduce that = g on ∂Ω.)
∂Ω ∂n ∂Ω ∂n
c) Prove that for any f ∈ L2 (Ω) and for any g ∈ L2 (∂Ω) there exists a unique weak solution
u ∈ Hm1 (Ω) to problem (9). Moreover, we have the estimate

kukH 1 (Ω) 6 C kf kL2 (Ω) + kgkL2 (∂Ω) .
Z
(Hint: Use the Lax-Milgram Theorem in Hm
1 (Ω) with the bilinear form B(u, v) = ∇u.∇v dx
Z Z Ω
and the functional Φ(v) = f v dx + gv dσ. Use the Poincaré inequality to prove that B is
Ω ∂Ω
coercive.)
1 1
It extends to a continuous mapping from H 1 (Ω) to H 2 (∂Ω), but this stronger result is not needed here.

30
 N  
∂ ∂u
Exercise 5. Consider the operator Au = − . The conormal derivative
X
ai,j (x)
∂xi ∂xj
i,j=1
∂u
associated with the operator A, denoted , is
∂nA
N
∂u ∂u
at any point x ∈ ∂Ω, where n is the outer normal to ∂Ω.
X
(x) = ai,j (x) (x)ni
∂nA ∂xj
i,j=1

Consider the boundary value problem

in Ω


 Au + c(x)u = f

(12)
∂u
=g on ∂Ω.



∂nA

We assume throughout that Ω is bounded with smooth boundary, ai,j ∈ L∞ (Ω) and A is
uniformly elliptic in Ω.
a) Suppose that c ∈ L∞ (Ω) and c(x) > κ > 0 for a.e. x ∈ Ω. Give a convenient denition
for weak solutions of (12) in H 1 (Ω). Then prove that for any f ∈ L2 (Ω) and any g ∈ L2 (∂Ω)
there exists a unique weak solution u ∈ H 1 (Ω) to problem (12) and we have the estimate


kukH 1 (Ω) 6 C kf kL2 (Ω) + kgkL2 (∂Ω) .

b) Suppose that c = 0 and ai,j ∈ C 1 (Ω). Show that (12) may have solutions only if the
compatibility condition (10) is satised. If this condition holds, dene the notion of weak
1 (Ω). Then prove that for any f ∈ L2 (Ω) and any g ∈ L2 (∂Ω) there exists
solution to (12) in Hm
a unique weak solution u ∈ Hm 1 (Ω) to problem (12) and the estimate in part (a) holds true.

(Hint: proceed as in the previous exercise.)

References

[1] J.-M. Bony, Cours d'Analyse - Théorie des distributions et analyse de Fourier, Éditions
de l'École Polytechnique, Palaiseau, 2011.
[2] H. Brezis, Analyse Fonctionnelle, Masson, Paris, 1983.
[3] H. Brezis, Functional analysis, Sobolev spaces and partial dierential equations, Springer
(Universitext), 2010.
[4] L. C. Evans, Partial Dierential Equations, 2nd edition, AMS (Graduate Studies in
Mathematics), Providence, Rhode Island, 2010.
[5] J. Smoller, Shock Waves and Reaction-Diusion Equations, 2nd edition, Springer-Verlag,
1994.
[6] W. Rudin, Functional Analysis, 2nd edition, McGraw-Hill, 1991.
[7] R. Strichartz, A guide to distribution theory and Fourier transforms, CRC Press, 1994.

31